File size: 109,417 Bytes
29b445b 26d8df8 29b445b 26d8df8 29b445b 26d8df8 29b445b 26d8df8 29b445b 26d8df8 29b445b 26d8df8 29b445b 26d8df8 29b445b 26d8df8 29b445b 26d8df8 29b445b 26d8df8 29b445b 26d8df8 29b445b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 |
#!/usr/bin/env python3
"""
Image Tagger Application
A Streamlit web app for tagging images using an AI model.
"""
import streamlit as st
import os
import sys
import traceback
import tempfile
import time
import platform
import subprocess
import webbrowser
import glob
import numpy as np
import matplotlib.pyplot as plt
import io
import base64
from matplotlib.colors import LinearSegmentedColormap
from PIL import Image
from pathlib import Path
# Add parent directory to path to allow importing from utils
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
# Import utilities
from utils.model_loader import load_exported_model, is_windows, check_flash_attention
from utils.image_processing import process_image, batch_process_images
from utils.file_utils import save_tags_to_file, get_default_save_locations
from utils.ui_components import display_progress_bar, show_example_images, display_batch_results
from utils.onnx_processing import batch_process_images_onnx
# Define the model directory
MODEL_DIR = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "model")
print(f"Using model directory: {MODEL_DIR}")
# Define threshold profile descriptions and explanations
threshold_profile_descriptions = {
"Micro Optimized": "Maximizes micro-averaged F1 score (best for dominant classes). Optimal for overall prediction quality.",
"Macro Optimized": "Maximizes macro-averaged F1 score (equal weight to all classes). Better for balanced performance across all tags.",
"Balanced": "Provides a trade-off between precision and recall with moderate thresholds. Good general-purpose setting.",
"High Precision": "Uses higher thresholds to prioritize accuracy over recall. Fewer but more confident predictions.",
"High Recall": "Uses lower thresholds to capture more potential tags at the expense of accuracy. More comprehensive tagging.",
"Overall": "Uses a single threshold value across all categories. Simplest approach for consistent behavior.",
"Weighted": "Uses thresholds weighted by category importance. Better balance for tags that matter most.",
"Category-specific": "Uses different optimal thresholds for each category. Best for fine-tuning results."
}
threshold_profile_explanations = {
"Micro Optimized": """
### Micro Optimized Profile
**Technical definition**: Maximizes micro-averaged F1 score, which calculates metrics globally across all predictions.
**When to use**: When you want the best overall accuracy, especially for common tags and dominant categories.
**Effects**:
- Optimizes performance for the most frequent tags
- Gives more weight to categories with many examples (like 'character' and 'general')
- Provides higher precision in most common use cases
**Threshold value**: Approximately 0.33 (optimized on validation data)
**Performance metrics**:
- Micro F1: ~0.62
- Macro F1: ~0.35
- Precision: ~0.63
- Recall: ~0.60
""",
"Macro Optimized": """
### Macro Optimized Profile
**Technical definition**: Maximizes macro-averaged F1 score, which gives equal weight to all categories regardless of size.
**When to use**: When balanced performance across all categories is important, including rare tags.
**Effects**:
- More balanced performance across all tag categories
- Better at detecting rare or unusual tags
- Generally has lower thresholds than micro-optimized
**Threshold value**: Approximately 0.19-0.21 (optimized on validation data)
**Performance metrics**:
- Micro F1: ~0.49
- Macro F1: ~0.41
- Precision: ~0.37
- Recall: ~0.53
""",
"Balanced": """
### Balanced Profile
**Technical definition**: Provides a compromise between precision and recall with moderate thresholds.
**When to use**: For general-purpose tagging when you don't have specific recall or precision requirements.
**Effects**:
- Good middle ground between precision and recall
- Works well for most common use cases
- Default choice for most users
**Threshold value**: Approximately 0.26 (optimized on validation data)
**Performance metrics**:
- Micro F1: ~0.59
- Macro F1: ~0.39
- Precision: ~0.51
- Recall: ~0.70
""",
"High Precision": """
### High Precision Profile
**Technical definition**: Uses higher thresholds to prioritize precision (correctness) over recall (coverage).
**When to use**: When you need high confidence in the tags that are returned and prefer to miss tags rather than include incorrect ones.
**Effects**:
- Much higher precision (84-97% of returned tags are correct)
- Lower recall (only captures 35-60% of relevant tags)
- Returns fewer tags overall, but with higher confidence
**Threshold value**: 0.50 (optimized for precision on validation data)
**Performance metrics**:
- Micro F1: ~0.50
- Macro F1: ~0.22
- Precision: ~0.84
- Recall: ~0.35
""",
"High Recall": """
### High Recall Profile
**Technical definition**: Uses lower thresholds to prioritize recall (coverage) over precision (correctness).
**When to use**: When you want to capture as many potential tags as possible and don't mind some incorrect suggestions.
**Effects**:
- Very high recall (captures 90%+ of relevant tags)
- Much lower precision (only 18-49% of returned tags may be correct)
- Returns many more tags, including less confident ones
**Threshold value**: 0.10-0.12 (optimized for recall on validation data)
**Performance metrics**:
- Micro F1: ~0.30
- Macro F1: ~0.35
- Precision: ~0.18
- Recall: ~0.90
""",
"Overall": """
### Overall Profile
**Technical definition**: Uses a single threshold value across all categories.
**When to use**: When you want consistent behavior across all categories and a simple approach.
**Effects**:
- Consistent tagging threshold for all categories
- Simpler to understand than category-specific thresholds
- User-adjustable with a single slider
**Default threshold value**: 0.35 (uses "balanced" threshold by default)
**Note**: The threshold value is user-adjustable with the slider below.
""",
"Weighted": """
### Weighted Profile
**Technical definition**: Uses thresholds weighted by category importance.
**When to use**: When you want different sensitivity for different categories based on their importance.
**Effects**:
- More important categories (like character and copyright) get optimized thresholds
- Less important categories get adjusted thresholds based on their contribution
- Better balance for the tags that matter most
**Default threshold values**: Varies by category (based on importance weighting)
**Note**: This uses pre-calculated optimal thresholds that can't be adjusted directly.
""",
"Category-specific": """
### Category-specific Profile
**Technical definition**: Uses different optimal thresholds for each category, allowing fine-tuning.
**When to use**: When you want to customize tagging sensitivity for different categories.
**Effects**:
- Each category has its own independent threshold
- Full control over category sensitivity
- Best for fine-tuning results when some categories need different treatment
**Default threshold values**: Starts with balanced thresholds for each category
**Note**: Use the category sliders below to adjust thresholds for individual categories.
"""
}
def get_profile_metrics(thresholds, profile_name, model_type="refined"):
"""
Extract metrics for the given profile from the thresholds dictionary
Args:
thresholds: The thresholds dictionary
profile_name: Name of the profile (micro_opt, macro_opt, balanced, etc.)
model_type: 'initial' or 'refined'
Returns:
Dictionary with metrics or None if not found
"""
profile_key = None
# Map UI-friendly names to internal keys
if profile_name == "Micro Optimized":
profile_key = "micro_opt"
elif profile_name == "Macro Optimized":
profile_key = "macro_opt"
elif profile_name == "Balanced":
profile_key = "balanced"
elif profile_name == "High Precision":
profile_key = "high_precision"
elif profile_name == "High Recall":
profile_key = "high_recall"
# For overall/weighted/category-specific, we're using the balanced profile metrics
elif profile_name in ["Overall", "Weighted", "Category-specific"]:
profile_key = "balanced"
# Handle the new JSON structure with "initial" and "refined" top-level keys
if "initial" in thresholds and "refined" in thresholds:
# Get the appropriate model type
model_type_key = model_type # Use the passed in model_type
# Make sure the model_type_key is valid
if model_type_key not in thresholds:
model_type_key = "refined" if "refined" in thresholds else "initial"
# Check if the profile exists
if "overall" in thresholds[model_type_key] and profile_key in thresholds[model_type_key]["overall"]:
return thresholds[model_type_key]["overall"][profile_key]
else:
# Fallback to the old structure
if "overall" in thresholds and profile_key in thresholds["overall"]:
return thresholds["overall"][profile_key]
return None
def on_threshold_profile_change():
"""
Handle threshold profile changes to ensure smooth transitions between modes
and preserve user customizations
"""
# Get the new profile
new_profile = st.session_state.threshold_profile
# Initialize active_threshold and active_category_thresholds based on profile
if hasattr(st.session_state, 'thresholds') and hasattr(st.session_state, 'settings'):
# Create category thresholds if they don't exist
if st.session_state.settings['active_category_thresholds'] is None:
st.session_state.settings['active_category_thresholds'] = {}
# Get existing thresholds
current_thresholds = st.session_state.settings['active_category_thresholds']
# Get model type for accessing thresholds - handle the new JSON structure
if "initial" in st.session_state.thresholds and "refined" in st.session_state.thresholds:
model_type_key = "refined" if hasattr(st.session_state, 'model_type') and st.session_state.model_type == "full" else "initial"
# Make sure the model_type_key is valid
if model_type_key not in st.session_state.thresholds:
model_type_key = "refined" if "refined" in st.session_state.thresholds else "initial"
else:
# Use None for old structure
model_type_key = None
# Map profile display names to internal keys
profile_key = None
if new_profile == "Micro Optimized":
profile_key = "micro_opt"
elif new_profile == "Macro Optimized":
profile_key = "macro_opt"
elif new_profile == "Balanced":
profile_key = "balanced"
elif new_profile == "High Precision":
profile_key = "high_precision"
elif new_profile == "High Recall":
profile_key = "high_recall"
# For specialized profiles, update thresholds from the thresholds dictionary
if profile_key:
# Set overall threshold based on JSON structure
if model_type_key is not None: # New structure
if "overall" in st.session_state.thresholds[model_type_key] and profile_key in st.session_state.thresholds[model_type_key]["overall"]:
st.session_state.settings['active_threshold'] = st.session_state.thresholds[model_type_key]["overall"][profile_key]["threshold"]
else: # Old structure
if "overall" in st.session_state.thresholds and profile_key in st.session_state.thresholds["overall"]:
st.session_state.settings['active_threshold'] = st.session_state.thresholds["overall"][profile_key]["threshold"]
# Set category thresholds based on JSON structure
for category in st.session_state.categories:
if model_type_key is not None: # New structure
if "categories" in st.session_state.thresholds[model_type_key] and category in st.session_state.thresholds[model_type_key]["categories"]:
if profile_key in st.session_state.thresholds[model_type_key]["categories"][category]:
current_thresholds[category] = st.session_state.thresholds[model_type_key]["categories"][category][profile_key]["threshold"]
else:
# Fallback to overall threshold if profile not found for this category
current_thresholds[category] = st.session_state.settings['active_threshold']
else: # Old structure
if "categories" in st.session_state.thresholds and category in st.session_state.thresholds["categories"]:
if profile_key in st.session_state.thresholds["categories"][category]:
current_thresholds[category] = st.session_state.thresholds["categories"][category][profile_key]["threshold"]
else:
current_thresholds[category] = st.session_state.settings['active_threshold']
# For "Overall" profile, reset to use just the overall threshold
elif new_profile == "Overall":
# Use the balanced threshold for Overall profile based on JSON structure
if model_type_key is not None: # New structure
if "overall" in st.session_state.thresholds[model_type_key] and "balanced" in st.session_state.thresholds[model_type_key]["overall"]:
st.session_state.settings['active_threshold'] = st.session_state.thresholds[model_type_key]["overall"]["balanced"]["threshold"]
else: # Old structure
if "overall" in st.session_state.thresholds and "balanced" in st.session_state.thresholds["overall"]:
st.session_state.settings['active_threshold'] = st.session_state.thresholds["overall"]["balanced"]["threshold"]
# Clear any category-specific overrides
st.session_state.settings['active_category_thresholds'] = {}
# For "Weighted" profile, use weighted thresholds
elif new_profile == "Weighted":
# Use the balanced threshold as base for Weighted profile based on JSON structure
if model_type_key is not None: # New structure
if "overall" in st.session_state.thresholds[model_type_key] and "balanced" in st.session_state.thresholds[model_type_key]["overall"]:
st.session_state.settings['active_threshold'] = st.session_state.thresholds[model_type_key]["overall"]["balanced"]["threshold"]
else: # Old structure
if "overall" in st.session_state.thresholds and "balanced" in st.session_state.thresholds["overall"]:
st.session_state.settings['active_threshold'] = st.session_state.thresholds["overall"]["balanced"]["threshold"]
# Get weighted thresholds if they exist, otherwise use balanced
if model_type_key is not None: # New structure
if "weighted" in st.session_state.thresholds[model_type_key]:
weighted_thresholds = st.session_state.thresholds[model_type_key]["weighted"]
for category in st.session_state.categories:
if category in weighted_thresholds:
current_thresholds[category] = weighted_thresholds[category]
else:
# Fallback to balanced threshold for categories not in weighted
if "categories" in st.session_state.thresholds[model_type_key] and category in st.session_state.thresholds[model_type_key]["categories"]:
if "balanced" in st.session_state.thresholds[model_type_key]["categories"][category]:
current_thresholds[category] = st.session_state.thresholds[model_type_key]["categories"][category]["balanced"]["threshold"]
else:
current_thresholds[category] = st.session_state.settings['active_threshold']
else:
current_thresholds[category] = st.session_state.settings['active_threshold']
else: # Old structure
if "weighted" in st.session_state.thresholds:
weighted_thresholds = st.session_state.thresholds["weighted"]
for category in st.session_state.categories:
if category in weighted_thresholds:
current_thresholds[category] = weighted_thresholds[category]
else:
# Fallback to balanced threshold
if "categories" in st.session_state.thresholds and category in st.session_state.thresholds["categories"]:
if "balanced" in st.session_state.thresholds["categories"][category]:
current_thresholds[category] = st.session_state.thresholds["categories"][category]["balanced"]["threshold"]
else:
current_thresholds[category] = st.session_state.settings['active_threshold']
else:
current_thresholds[category] = st.session_state.settings['active_threshold']
# For "Category-specific", initialize with balanced thresholds
elif new_profile == "Category-specific":
# Use the balanced threshold as base for Category-specific based on JSON structure
if model_type_key is not None: # New structure
if "overall" in st.session_state.thresholds[model_type_key] and "balanced" in st.session_state.thresholds[model_type_key]["overall"]:
st.session_state.settings['active_threshold'] = st.session_state.thresholds[model_type_key]["overall"]["balanced"]["threshold"]
else: # Old structure
if "overall" in st.session_state.thresholds and "balanced" in st.session_state.thresholds["overall"]:
st.session_state.settings['active_threshold'] = st.session_state.thresholds["overall"]["balanced"]["threshold"]
# Initialize with balanced thresholds for each category
for category in st.session_state.categories:
if model_type_key is not None: # New structure
if "categories" in st.session_state.thresholds[model_type_key] and category in st.session_state.thresholds[model_type_key]["categories"]:
if "balanced" in st.session_state.thresholds[model_type_key]["categories"][category]:
current_thresholds[category] = st.session_state.thresholds[model_type_key]["categories"][category]["balanced"]["threshold"]
else:
current_thresholds[category] = st.session_state.settings['active_threshold']
else:
current_thresholds[category] = st.session_state.settings['active_threshold']
else: # Old structure
if "categories" in st.session_state.thresholds and category in st.session_state.thresholds["categories"]:
if "balanced" in st.session_state.thresholds["categories"][category]:
current_thresholds[category] = st.session_state.thresholds["categories"][category]["balanced"]["threshold"]
else:
current_thresholds[category] = st.session_state.settings['active_threshold']
else:
current_thresholds[category] = st.session_state.settings['active_threshold']
def create_micro_macro_comparison():
"""
Creates a visual explanation of micro vs macro optimization
Returns:
HTML for the visualization
"""
html = """
<style>
.optimization-container {
font-family: sans-serif;
margin: 20px 0;
}
.optimization-row {
display: flex;
margin-bottom: 15px;
}
.optimization-col {
flex: 1;
padding: 15px;
border-radius: 8px;
margin: 0 5px;
}
.optimization-col h3 {
margin-top: 0;
font-size: 18px;
}
.optimization-col p {
font-size: 14px;
line-height: 1.5;
}
.micro-col {
background-color: #e6f3ff;
border: 1px solid #99ccff;
}
.macro-col {
background-color: #fff0e6;
border: 1px solid #ffcc99;
}
.tag-example {
display: inline-block;
padding: 3px 8px;
margin: 3px;
border-radius: 12px;
font-size: 12px;
}
.tag-common {
background-color: #4CAF50;
color: white;
}
.tag-rare {
background-color: #9C27B0;
color: white;
}
.comparison-table {
width: 100%;
border-collapse: collapse;
margin-top: 15px;
}
.comparison-table th, .comparison-table td {
border: 1px solid #ddd;
padding: 8px;
text-align: left;
}
.comparison-table th {
background-color: #f2f2f2;
}
</style>
"""
return html
def apply_thresholds(all_probs, threshold_profile, active_threshold, active_category_thresholds, min_confidence, selected_categories):
"""
Apply thresholds to raw probabilities and return filtered tags
Args:
all_probs: Dictionary with all probabilities organized by category
threshold_profile: Current threshold profile
active_threshold: Overall threshold value
active_category_thresholds: Dictionary of category-specific thresholds
min_confidence: Minimum confidence to include
selected_categories: Dictionary of selected categories
Returns:
tags: Dictionary of filtered tags above threshold by category
all_tags: List of all tags above threshold
"""
# Apply thresholds to each category
tags = {}
all_tags = []
for category, cat_probs in all_probs.items():
# Get the appropriate threshold for this category
# For Overall and Weighted profiles, respect per-category overrides if available
threshold = active_category_thresholds.get(category, active_threshold) if active_category_thresholds else active_threshold
# Filter tags above threshold
tags[category] = [(tag, prob) for tag, prob in cat_probs if prob >= threshold]
# Add to all_tags if selected
if selected_categories.get(category, True):
for tag, prob in tags[category]:
all_tags.append(tag)
return tags, all_tags
def image_tagger_app():
"""Main Streamlit application for image tagging."""
st.set_page_config(layout="wide", page_title="Camie Tagger", page_icon="🖼️")
st.title("Image Tagging Interface")
st.markdown("---")
# Check platform and Flash Attention
windows_system = is_windows()
flash_attn_installed = check_flash_attention()
if 'settings' not in st.session_state:
st.session_state.settings = {
'show_all_tags': False,
'compact_view': True,
'min_confidence': 0.01,
'threshold_profile': "Balanced",
'active_threshold': 0.35,
'active_category_thresholds': None,
'selected_categories': {},
'replace_underscores': False # Added new setting
}
# Initialize show_profile_help state
st.session_state.show_profile_help = False
# Define default threshold values (initialized here to avoid errors)
default_threshold_values = {
'overall': 0.35,
'weighted': 0.35,
'category_thresholds': {},
'high_precision_thresholds': {},
'high_recall_thresholds': {}
}
# Session state initialization for model
if 'model_loaded' not in st.session_state:
st.session_state.model_loaded = False
st.session_state.model = None
st.session_state.thresholds = None
st.session_state.metadata = None
# Check if ONNX model files exist
onnx_model_path = os.path.join(os.path.dirname(MODEL_DIR), "model_initial.onnx")
onnx_metadata_path = os.path.join(os.path.dirname(MODEL_DIR), "model_initial_metadata.json")
onnx_available = os.path.exists(onnx_model_path) and os.path.exists(onnx_metadata_path)
# Default to ONNX if available, otherwise fallback to initial_only on Windows or full elsewhere
if onnx_available:
st.session_state.model_type = "onnx"
else:
st.session_state.model_type = "initial_only" if windows_system else "full"
# Sidebar for model selection and information
with st.sidebar:
st.header("Model Selection")
# Check if ONNX model files exist
onnx_model_path = os.path.join(os.path.dirname(MODEL_DIR), "model_initial.onnx")
onnx_metadata_path = os.path.join(os.path.dirname(MODEL_DIR), "model_initial_metadata.json")
onnx_available = os.path.exists(onnx_model_path) and os.path.exists(onnx_metadata_path)
# Define model options, including ONNX if available
model_options = [
"Refined (Tag Embeddings)",
"Initial (Base Model)"
]
# Add ONNX option if available
if onnx_available:
model_options.append("ONNX Accelerated (Fastest)")
# Determine the default index for model selection
if st.session_state.model_type == "onnx" and onnx_available:
default_index = 2 # ONNX is first priority when available
elif windows_system or st.session_state.model_type == "initial_only":
default_index = 1 # Initial Only is second priority, default on Windows
else:
default_index = 0 # Full model is last priority
# Model type selection with radio buttons
model_type = st.radio(
"Select Model Type:",
model_options,
index=min(default_index, len(model_options)-1), # Ensure index is valid
help="""
Full Model: Uses both initial and refined predictions for highest accuracy (requires more VRAM)
Initial Only: Uses only the initial classifier, reducing VRAM usage at slight quality cost
ONNX Accelerated: Optimized for inference speed, best for batch processing (if available)
"""
)
# Convert selection to internal model type
if model_type == "Full Model (Best Quality)":
selected_model_type = "full"
elif model_type == "ONNX Accelerated (Fastest)":
selected_model_type = "onnx"
# Store ONNX paths in session state for later use
st.session_state.onnx_model_path = onnx_model_path
st.session_state.onnx_metadata_path = onnx_metadata_path
else: # "Initial Only (Lower VRAM)"
selected_model_type = "initial_only"
# If the model type has changed, we need to reload the model
if selected_model_type != st.session_state.model_type:
st.session_state.model_loaded = False
st.session_state.model_type = selected_model_type
# Add Windows warning if relevant
if windows_system and selected_model_type=="full":
st.warning("""
### Windows Compatibility Note
The refined model requires Flash Attention which is difficult to install on Windows.
For Windows users, I recommend using the "Initial Only" or ONNX Accelerated model which:
- Does not require Flash Attention
- Uses less memory
- Provides very close to full prediction quality (check performance notes on HF)
""")
# Add a button to reload the model with current settings
if st.button("Reload Model") and st.session_state.model_loaded:
st.session_state.model_loaded = False
st.info("Reloading model...")
# Try to load the model when not loaded
if not st.session_state.model_loaded:
try:
with st.spinner(f"Loading {st.session_state.model_type} model..."):
if st.session_state.model_type == "onnx":
# Load ONNX model and metadata
import json
import onnxruntime as ort
try:
# Check ONNX providers (for info display)
providers = ort.get_available_providers()
gpu_available = any('GPU' in provider for provider in providers)
# Store provider info in session state
st.session_state.onnx_providers = providers
st.session_state.onnx_gpu_available = gpu_available
# Load metadata
with open(st.session_state.onnx_metadata_path, 'r') as f:
metadata = json.load(f)
# Load thresholds from a separate file or use defaults
thresholds_path = os.path.join(MODEL_DIR, "thresholds.json")
if os.path.exists(thresholds_path):
with open(thresholds_path, 'r') as f:
thresholds = json.load(f)
else:
# If no thresholds file, extract from metadata if available
if 'thresholds' in metadata:
thresholds = metadata['thresholds']
else:
# Use default thresholds if not available
thresholds = {
'overall': {'balanced': {'threshold': 0.35}},
'weighted': {'f1': {'threshold': 0.35}},
'categories': {}
}
# Build category thresholds if necessary
if 'tag_to_category' in metadata:
categories = set(metadata['tag_to_category'].values())
thresholds['categories'] = {
cat: {
'balanced': {'threshold': 0.35},
'high_precision': {'threshold': 0.45},
'high_recall': {'threshold': 0.25}
} for cat in categories
}
# Set device info for display
device = "ONNX Runtime" + (" (GPU)" if gpu_available else " (CPU)")
param_dtype = "float32" # Most ONNX models use float32
# Store information in session state
st.session_state.model = None # No PyTorch model for ONNX
st.session_state.device = device
st.session_state.param_dtype = param_dtype
st.session_state.thresholds = thresholds
st.session_state.metadata = metadata
st.session_state.model_loaded = True
# Get categories from metadata
categories = list(set(metadata['tag_to_category'].values()))
st.session_state.categories = categories
# Initialize selected categories if needed
if not st.session_state.settings['selected_categories']:
st.session_state.settings['selected_categories'] = {cat: True for cat in categories}
except Exception as e:
st.error(f"Error loading ONNX model: {str(e)}")
st.info(f"Make sure the ONNX model and metadata files exist at: {st.session_state.onnx_model_path} and {st.session_state.onnx_metadata_path}")
st.code(traceback.format_exc())
st.stop()
else:
# Load PyTorch model as before
model, thresholds, metadata = load_exported_model(
MODEL_DIR,
model_type=st.session_state.model_type
)
# Extract device and precision info
device = next(model.parameters()).device
param_dtype = next(model.parameters()).dtype
# Store model in session state for PyTorch models
st.session_state.model = model
# Common code for all model types
# Get available categories
categories = list(set(metadata['tag_to_category'].values()))
# Initialize selected categories (all selected by default)
if not st.session_state.settings['selected_categories']:
st.session_state.settings['selected_categories'] = {cat: True for cat in categories}
# Store common info in session state
st.session_state.device = device
st.session_state.param_dtype = param_dtype
st.session_state.thresholds = thresholds
st.session_state.metadata = metadata
st.session_state.model_loaded = True
st.session_state.categories = categories
# Debug: Print loaded thresholds to verify they're loaded correctly
print("Loaded thresholds:", thresholds)
if "initial" in thresholds and "refined" in thresholds:
# Choose which model type to use as default
model_type_key = "refined" if st.session_state.model_type == "full" else "initial"
# Set overall threshold from the balanced profile
if "overall" in thresholds[model_type_key] and "balanced" in thresholds[model_type_key]["overall"]:
default_threshold_values['overall'] = thresholds[model_type_key]["overall"]["balanced"]["threshold"]
# Get weighted threshold if available
if "weighted" in thresholds[model_type_key] and "f1" in thresholds[model_type_key]["weighted"]:
default_threshold_values['weighted'] = thresholds[model_type_key]["weighted"]["f1"]["threshold"]
# Set category thresholds
if "categories" in thresholds[model_type_key]:
default_threshold_values['category_thresholds'] = {
cat: opt['balanced']['threshold']
for cat, opt in thresholds[model_type_key]["categories"].items()
}
# Set high precision and high recall thresholds
default_threshold_values['high_precision_thresholds'] = {
cat: opt['high_precision']['threshold']
for cat, opt in thresholds[model_type_key]["categories"].items()
}
default_threshold_values['high_recall_thresholds'] = {
cat: opt['high_recall']['threshold']
for cat, opt in thresholds[model_type_key]["categories"].items()
}
else:
# Fallback to the old structure for backward compatibility
if "overall" in thresholds and "balanced" in thresholds["overall"]:
default_threshold_values['overall'] = thresholds["overall"]["balanced"]["threshold"]
# Get weighted threshold if available
if "weighted" in thresholds and "f1" in thresholds["weighted"]:
default_threshold_values['weighted'] = thresholds["weighted"]["f1"]["threshold"]
# Set category thresholds
if "categories" in thresholds:
default_threshold_values['category_thresholds'] = {
cat: opt['balanced']['threshold']
for cat, opt in thresholds["categories"].items()
}
# Set high precision and high recall thresholds
default_threshold_values['high_precision_thresholds'] = {
cat: opt['high_precision']['threshold']
for cat, opt in thresholds["categories"].items()
}
default_threshold_values['high_recall_thresholds'] = {
cat: opt['high_recall']['threshold']
for cat, opt in thresholds["categories"].items()
}
# Update session state with current threshold values
# This part is crucial - store the values in session state
st.session_state.default_threshold_values = default_threshold_values
# Update active threshold with default values
if st.session_state.settings['threshold_profile'] == "Overall":
st.session_state.settings['active_threshold'] = default_threshold_values['overall']
elif st.session_state.settings['threshold_profile'] == "Weighted":
st.session_state.settings['active_threshold'] = default_threshold_values['weighted']
except Exception as e:
st.error(f"Error loading model: {str(e)}")
st.info(f"Looking for model in: {os.path.abspath(MODEL_DIR)}")
# Check for specific files
if st.session_state.model_type == "initial_only":
expected_model_paths = [
os.path.join(MODEL_DIR, "model_initial_only.pt"),
os.path.join(MODEL_DIR, "model_initial.pt")
]
if not any(os.path.exists(p) for p in expected_model_paths):
st.error(f"Initial-only model file not found. Checked: {', '.join(expected_model_paths)}")
st.info("Make sure you've exported both model types.")
else:
expected_model_paths = [
os.path.join(MODEL_DIR, "model_refined.pt"),
os.path.join(MODEL_DIR, "model.pt"),
os.path.join(MODEL_DIR, "model_full.pt")
]
if not any(os.path.exists(p) for p in expected_model_paths):
st.error(f"Full model file not found. Checked: {', '.join(expected_model_paths)}")
st.code(traceback.format_exc())
st.stop()
with st.sidebar:
st.header("Model Information")
if st.session_state.model_loaded:
# Show model type
if st.session_state.model_type == "onnx":
st.success("Using ONNX Accelerated Model")
# Show ONNX-specific info
if hasattr(st.session_state, 'onnx_gpu_available') and st.session_state.onnx_gpu_available:
st.write("Acceleration: GPU available")
else:
st.write("Acceleration: CPU only")
elif st.session_state.model_type == "full":
st.success("Using Full Model (Best Quality)")
# Show Flash Attention info for full model
if not flash_attn_installed and is_windows():
st.warning("Note: Flash Attention not available on Windows")
else:
st.success("Using Initial-Only Model (Lower VRAM)")
# Show common model info
st.write(f"Device: {st.session_state.device}")
st.write(f"Precision: {st.session_state.param_dtype}")
st.write(f"Total tags: {st.session_state.metadata['total_tags']}")
# Show categories in an expander
with st.expander("Available Categories"):
for category in sorted(st.session_state.categories):
st.write(f"- {category.capitalize()}")
# Add an expander with information
with st.expander("About this app"):
st.write("""
This app uses a trained image tagging model to analyze and tag images.
**Model Options**:
- **ONNX Accelerated (Fastest)**: Optimized for inference speed with minimal VRAM usage, ideal for batch processing
- **Refined Model (Tag Embeddings)**: Higher quality predictions using both initial and refined layers (uses more VRAM)
- **Initial Model (Base model)**: Reduced VRAM usage with slightly lower accuracy (good for systems with limited resources)
**Platform Notes**:
- **Windows Users**: ONNX Accelerated model is recommended for best performance
- **CUDA Support**: GPU acceleration is available for ONNX models if CUDA 12.x and cuDNN are installed
- **Linux Users**: The Refined Model with Flash Attention provides the best quality results
**Features**:
- Upload or select an image
- Process multiple images in batch mode with customizable batch size
- Choose from different threshold profiles
- Adjust category-specific thresholds
- View predictions organized by category
- Limit results to top N tags within each category
- Save tags to text files in various locations
- Export tags with consistent formatting for external use
- Fast batch processing
**Threshold profiles**:
- **Micro Optimized**: Optimizes micro-averaged F1 score (best for common tags)
- **Macro Optimized**: Optimizes macro-averaged F1 score (better for rare tags)
- **Balanced**: Provides a good balance of precision and recall
- **High Precision**: Prioritizes accuracy over recall
- **High Recall**: Captures more potential tags but may be less accurate
""")
with st.sidebar:
# Add separator for visual clarity
st.markdown("---")
# Support information
st.subheader("💡 Notes")
st.markdown("""
This tagger was trained on a subset of the available data and for limited epochs due to hardware limitations.
A more comprehensive model trained on the full 3+ million image dataset and many more epochs would provide:
- More recent characters and tags.
- Improved accuracy.
If you find this tool useful and would like to support future development:
""")
# Add Buy Me a Coffee button with Star of the City-like glow effect
st.markdown("""
<style>
@keyframes coffee-button-glow {
0% { box-shadow: 0 0 5px #FFD700; }
50% { box-shadow: 0 0 15px #FFD700; }
100% { box-shadow: 0 0 5px #FFD700; }
}
.coffee-button {
display: inline-block;
animation: coffee-button-glow 2s infinite;
border-radius: 5px;
transition: transform 0.3s ease;
}
.coffee-button:hover {
transform: scale(1.05);
}
</style>
<a href="https://buymeacoffee.com/camais" target="_blank" class="coffee-button">
<img src="https://cdn.buymeacoffee.com/buttons/v2/default-yellow.png"
alt="Buy Me A Coffee"
style="height: 45px; width: 162px; border-radius: 5px;" />
</a>
""", unsafe_allow_html=True)
st.markdown("""
Your support helps with:
- GPU costs for training
- Storage for larger datasets
- Development of new features
- Future projects
Thank you! 🙏
Full Details: https://huggingface.co/Camais03/camie-tagger
""")
with st.sidebar:
# Add game link at the bottom of the sidebar
st.markdown("---")
st.subheader("Try the Tag Collector Game!")
st.write("Test your tagging skills in our gamified version of the image tagger!")
if st.button("🎮 Launch Tag Collector Game", type="primary"):
# Get the current port to determine the game URL
current_port = os.environ.get("STREAMLIT_SERVER_PORT", "8501")
# The game should run on a different port (8502 if this is 8501, or vice versa)
game_port = "8502" if current_port == "8501" else "8501"
# Check if the game file exists
game_path = os.path.join(os.path.dirname(__file__), "tag_collector_game.py")
if os.path.exists(game_path):
# Launch the game in a new process
try:
# Determine streamlit path
if sys.platform == "win32":
streamlit_path = os.path.join("venv", "Scripts", "streamlit.exe")
else:
streamlit_path = os.path.join("venv", "bin", "streamlit")
if not os.path.exists(streamlit_path):
streamlit_path = "streamlit" # Fallback to global streamlit
# Build command to run game on different port
command = [streamlit_path, "run", game_path, "--server.port", game_port]
# Launch in background
if sys.platform == "win32":
subprocess.Popen(command, shell=True, creationflags=subprocess.CREATE_NEW_CONSOLE)
else:
subprocess.Popen(command)
# Open in browser
game_url = f"http://localhost:{game_port}"
webbrowser.open(game_url)
st.success(f"Launching Tag Collector Game!")
except Exception as e:
st.error(f"Failed to launch game: {str(e)}")
else:
st.error(f"Game file not found: {game_path}")
st.info("Make sure tag_collector_game.py is in the same directory as this app.")
# Main content area
col1, col2 = st.columns([1, 1.5])
# Column 1: Image upload and display
with col1:
st.header("Image")
# Add tabs for Upload, Examples, and Batch Processing
upload_tab, batch_tab = st.tabs(["Upload Image", "Batch Processing"])
image_path = None
with upload_tab:
uploaded_file = st.file_uploader("Choose an image...", type=["jpg", "jpeg", "png"])
if uploaded_file:
# Create a temporary file
with tempfile.NamedTemporaryFile(delete=False, suffix='.jpg') as tmp_file:
tmp_file.write(uploaded_file.getvalue())
image_path = tmp_file.name
# Store the original filename for saving tags
st.session_state.original_filename = uploaded_file.name
# Display the image
image = Image.open(uploaded_file)
st.image(image, use_container_width=True)
with batch_tab:
st.subheader("Batch Process Images")
st.write("Process multiple images from a folder and save tags to text files.")
# Folder selection
batch_folder = st.text_input("Enter folder path containing images:", "")
if st.button("Browse Folder..."):
# This is a dummy button since Streamlit doesn't have a native folder picker
st.info("Please type the folder path manually in the text input above.")
# Save location options
save_options = st.radio(
"Where to save tag files:",
["Same folder as images", "Custom location", "Default save folder"],
index=0
)
# Add batch size control
st.subheader("Performance Options")
batch_size_text = st.text_input(
"Batch size (images processed at once)",
value="4",
help="Higher values may improve processing speed but use more memory. Recommended: 4-16"
)
# Convert to integer with error handling
try:
batch_size = int(batch_size_text)
if batch_size < 1:
st.warning("Batch size must be at least 1. Using batch size of 1.")
batch_size = 1
except ValueError:
st.warning("Please enter a valid number for batch size. Using default batch size of 4.")
batch_size = 4
# Performance note
if batch_size > 8:
st.info(f"Using larger batch size ({batch_size}). If you encounter memory issues, try reducing this value.")
st.write("Set tag limits per category for batch processing:")
# Create a toggle for enabling category limits
enable_category_limits = st.checkbox("Limit tags per category in batch output", value=False)
# Initialize the category limit dictionary if it doesn't exist
if 'category_limits' not in st.session_state:
st.session_state.category_limits = {}
if enable_category_limits:
# Create a two-column layout for more compact display
limit_cols = st.columns(2)
# Add an explanation about the values
st.markdown("""
**Limit Values:**
* **-1** = No limit (include all tags)
* **0** = Exclude category entirely
* **N** (positive number) = Include only top N tags
""")
if hasattr(st.session_state, 'categories'):
# Create text inputs for each category
for i, category in enumerate(sorted(st.session_state.categories)):
col_idx = i % 2 # Alternate between columns
with limit_cols[col_idx]:
# Get current limit value (default to -1 for unlimited)
current_limit = st.session_state.category_limits.get(category, -1)
# Add a text input for this category
limit_text = st.text_input(
f"{category.capitalize()} (top N):",
value=str(current_limit),
key=f"limit_{category}",
help="-1 = no limit, 0 = exclude, N = top N tags"
)
# Convert to integer with error handling
try:
limit = int(limit_text)
if limit < -1:
st.warning(f"Limit for {category} must be -1 or greater. Using -1 (unlimited).")
limit = -1
except ValueError:
st.warning(f"Invalid limit for {category}. Using -1 (unlimited).")
limit = -1
# Display a clear indicator of what this setting means
if limit == -1:
st.caption(f"✅ Including all {category} tags")
elif limit == 0:
st.caption(f"❌ Excluding all {category} tags")
else:
st.caption(f"⚙️ Including top {limit} {category} tags")
# Store the limit in session state
st.session_state.category_limits[category] = limit
else:
st.info("Categories will be available after loading a model.")
else:
# Clear any existing limits if disabled
st.session_state.category_limits = {}
custom_save_dir = None
if save_options == "Custom location":
# Allow selecting a custom save location
if 'custom_folders' not in st.session_state:
st.session_state.custom_folders = get_default_save_locations()
custom_save_dir = st.selectbox(
"Select save location:",
options=st.session_state.custom_folders,
format_func=lambda x: os.path.basename(x) if os.path.basename(x) else x
)
# Allow adding a new folder
new_folder = st.text_input("Or enter a new folder path:", key="batch_new_folder")
if st.button("Add Folder", key="batch_add_folder") and new_folder:
if os.path.isdir(new_folder):
if new_folder not in st.session_state.custom_folders:
st.session_state.custom_folders.append(new_folder)
st.success(f"Added folder: {new_folder}")
st.rerun()
else:
st.info("This folder is already in the list.")
else:
try:
# Try to create the folder if it doesn't exist
os.makedirs(new_folder, exist_ok=True)
st.session_state.custom_folders.append(new_folder)
st.success(f"Created and added folder: {new_folder}")
st.rerun()
except Exception as e:
st.error(f"Could not create folder: {str(e)}")
# Check if folder exists and count images
if batch_folder and os.path.isdir(batch_folder):
# Count image files
image_extensions = ['*.jpg', '*.jpeg', '*.png']
image_files = []
for ext in image_extensions:
image_files.extend(glob.glob(os.path.join(batch_folder, ext)))
image_files.extend(glob.glob(os.path.join(batch_folder, ext.upper())))
# Use a set to remove duplicate files (Windows filesystems are case-insensitive)
if os.name == 'nt': # Windows
# Use lowercase paths for comparison on Windows
unique_paths = set()
unique_files = []
for file_path in image_files:
normalized_path = os.path.normpath(file_path).lower()
if normalized_path not in unique_paths:
unique_paths.add(normalized_path)
unique_files.append(file_path)
image_files = unique_files
total_images = len(image_files)
st.write(f"Found {total_images} image files in the folder.")
# Show first few images as thumbnails (more compact layout)
if image_files:
st.write("Sample images:")
num_preview = min(8, len(image_files))
thumbnail_cols = st.columns(4)
for i, img_path in enumerate(image_files[:num_preview]):
with thumbnail_cols[i % 4]:
try:
img = Image.open(img_path)
# Make thumbnails smaller for more compact view
st.image(img, width=80, caption=os.path.basename(img_path))
except:
st.write(f"Error loading {os.path.basename(img_path)}")
# Determine save directory based on user selection
if save_options == "Same folder as images":
save_dir = batch_folder
elif save_options == "Custom location":
save_dir = custom_save_dir
else: # Default save folder
app_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
save_dir = os.path.join(app_dir, "saved_tags")
os.makedirs(save_dir, exist_ok=True)
# Add a more prominent button for batch processing
st.markdown("---")
process_col1, process_col2 = st.columns([3, 1])
with process_col1:
st.write(f"Ready to process {total_images} images")
st.write(f"Tags will be saved to: **{save_dir}**")
with process_col2:
# Check if model is loaded
process_button_disabled = not st.session_state.model_loaded
# Add a prominent batch processing button
if st.button("🔄 Process All Images",
key="process_batch_btn",
use_container_width=True,
disabled=process_button_disabled,
type="primary"):
if not st.session_state.model_loaded:
st.error("Model not loaded. Please check the model settings.")
else:
with st.spinner("Processing images..."):
# Create progress bar
progress_bar = st.progress(0)
status_text = st.empty()
# Define progress callback
def update_progress(current, total, image_path):
if total > 0:
progress = min(current / total, 1.0)
progress_bar.progress(progress)
if image_path:
status_text.text(f"Processing {current}/{total}: {os.path.basename(image_path)}")
else:
status_text.text(f"Completed processing {current}/{total} images")
# Get settings for batch processing
curr_threshold_profile = st.session_state.settings['threshold_profile']
curr_active_threshold = st.session_state.settings['active_threshold']
curr_active_category_thresholds = st.session_state.settings['active_category_thresholds']
curr_min_confidence = st.session_state.settings['min_confidence']
# Get category limits if enabled
curr_category_limits = None
if 'category_limits' in st.session_state and enable_category_limits:
curr_category_limits = st.session_state.category_limits
# Print debugging info in a more structured way
if curr_category_limits:
st.write("Category limit settings:")
# Group by type for cleaner display
excluded = []
limited = []
unlimited = []
for cat, limit in sorted(curr_category_limits.items()):
if limit == 0:
excluded.append(cat)
elif limit > 0:
limited.append(f"{cat}: top {limit}")
else: # limit == -1
unlimited.append(cat)
# Show in a more structured way
if excluded:
st.write("❌ Excluded categories: " + ", ".join(excluded))
if limited:
st.write("⚙️ Limited categories: " + ", ".join(limited))
if unlimited:
st.write("✅ Unlimited categories: " + ", ".join(unlimited))
if not excluded and not limited:
st.write("No limits set (all categories included)")
# Then when you call the batch processing functions, pass this parameter:
if st.session_state.model_type == "onnx":
# Use ONNX batch processing
batch_results = batch_process_images_onnx(
folder_path=batch_folder,
model_path=st.session_state.onnx_model_path,
metadata_path=st.session_state.onnx_metadata_path,
threshold_profile=curr_threshold_profile,
active_threshold=curr_active_threshold,
active_category_thresholds=curr_active_category_thresholds,
save_dir=save_dir,
progress_callback=update_progress,
min_confidence=curr_min_confidence,
batch_size=batch_size,
category_limits=curr_category_limits
)
else:
# Use standard PyTorch processing
batch_results = batch_process_images(
folder_path=batch_folder,
model=st.session_state.model,
thresholds=st.session_state.thresholds,
metadata=st.session_state.metadata,
threshold_profile=curr_threshold_profile,
active_threshold=curr_active_threshold,
active_category_thresholds=curr_active_category_thresholds,
save_dir=save_dir,
progress_callback=update_progress,
min_confidence=curr_min_confidence,
batch_size=batch_size,
category_limits=st.session_state.category_limits if enable_category_limits else None
)
# Display results
display_batch_results(batch_results)
# If model is not loaded, show a hint
if not st.session_state.model_loaded:
st.warning("Please load a model before processing images.")
else:
st.warning("No image files found in the selected folder.")
elif batch_folder:
st.error(f"Folder not found: {batch_folder}")
# Column 2: Threshold controls and predictions
with col2:
st.header("Tagging Controls")
# Define all available profiles
all_profiles = [
"Micro Optimized",
"Macro Optimized",
"Balanced",
"High Precision",
"High Recall",
"Overall",
"Weighted",
"Category-specific"
]
# Define a default index based on what's most similar to the current setting
default_index = 2 # Default to "Balanced"
if "threshold_profile" in st.session_state.settings:
# If there's an existing selection, try to match it
existing_profile = st.session_state.settings['threshold_profile']
if existing_profile in all_profiles:
default_index = all_profiles.index(existing_profile)
# Map old profile names to new ones
elif existing_profile == "Overall":
default_index = all_profiles.index("Overall")
elif existing_profile == "Weighted":
default_index = all_profiles.index("Weighted")
elif existing_profile == "Category-specific":
default_index = all_profiles.index("Category-specific")
elif existing_profile == "High Precision":
default_index = all_profiles.index("High Precision")
elif existing_profile == "High Recall":
default_index = all_profiles.index("High Recall")
# Create the profile selection UI with help button
profile_col1, profile_col2 = st.columns([3, 1])
with profile_col1:
# Create the profile dropdown
threshold_profile = st.selectbox(
"Select threshold profile",
options=all_profiles,
index=default_index,
key="threshold_profile",
on_change=on_threshold_profile_change
)
with profile_col2:
# Add a help button that expands to show detailed information
if st.button("ℹ️ Help", key="profile_help"):
st.session_state.show_profile_help = not st.session_state.get('show_profile_help', False)
# Display the help text if the button was clicked
if st.session_state.get('show_profile_help', False):
st.markdown(threshold_profile_explanations[threshold_profile])
else:
# Just show the description
st.info(threshold_profile_descriptions[threshold_profile])
# Get profile metrics from thresholds
if st.session_state.model_loaded:
# Try to get metrics for the selected profile
model_type = "refined" if st.session_state.model_type == "full" else "initial"
metrics = get_profile_metrics(st.session_state.thresholds, threshold_profile, model_type)
if metrics:
# Create metrics display
metrics_cols = st.columns(3)
with metrics_cols[0]:
# Display threshold
threshold_value = metrics.get("threshold", 0.35)
st.metric("Threshold", f"{threshold_value:.3f}")
with metrics_cols[1]:
# Display micro F1
micro_f1 = metrics.get("micro_f1", metrics.get("micro_precision", 0))
st.metric("Micro F1", f"{micro_f1:.3f}" if micro_f1 else "N/A")
# Show precision if available
precision = metrics.get("precision", metrics.get("micro_precision", 0))
if precision:
st.metric("Precision", f"{precision:.3f}")
with metrics_cols[2]:
# Display macro F1
macro_f1 = metrics.get("macro_f1", 0)
st.metric("Macro F1", f"{macro_f1:.3f}" if macro_f1 else "N/A")
# Show recall if available
recall = metrics.get("recall", metrics.get("micro_recall", 0))
if recall:
st.metric("Recall", f"{recall:.3f}")
# Initialize thresholds based on the selected profile
active_threshold = None
active_category_thresholds = {}
# Handle the new JSON structure
if st.session_state.model_loaded:
# Determine which model type key to use
if "initial" in st.session_state.thresholds and "refined" in st.session_state.thresholds:
model_type_key = "refined" if st.session_state.model_type == "full" else "initial"
# Make sure the model_type_key is valid
if model_type_key not in st.session_state.thresholds:
model_type_key = "refined" if "refined" in st.session_state.thresholds else "initial"
else:
model_type_key = None # Old structure - access thresholds directly
# Map profile display names to internal keys
profile_key = None
if threshold_profile == "Micro Optimized":
profile_key = "micro_opt"
elif threshold_profile == "Macro Optimized":
profile_key = "macro_opt"
elif threshold_profile == "Balanced":
profile_key = "balanced"
elif threshold_profile == "High Precision":
profile_key = "high_precision"
elif threshold_profile == "High Recall":
profile_key = "high_recall"
# For specialized profiles, get thresholds from the thresholds dictionary
if profile_key:
# Get overall threshold
if model_type_key is not None: # New structure
if "overall" in st.session_state.thresholds[model_type_key] and profile_key in st.session_state.thresholds[model_type_key]["overall"]:
active_threshold = st.session_state.thresholds[model_type_key]["overall"][profile_key]["threshold"]
else: # Old structure
if "overall" in st.session_state.thresholds and profile_key in st.session_state.thresholds["overall"]:
active_threshold = st.session_state.thresholds["overall"][profile_key]["threshold"]
# Get category thresholds
for category in st.session_state.categories:
if model_type_key is not None: # New structure
if "categories" in st.session_state.thresholds[model_type_key] and category in st.session_state.thresholds[model_type_key]["categories"]:
if profile_key in st.session_state.thresholds[model_type_key]["categories"][category]:
active_category_thresholds[category] = st.session_state.thresholds[model_type_key]["categories"][category][profile_key]["threshold"]
else:
# Fallback to overall threshold if profile not found for this category
active_category_thresholds[category] = active_threshold
else:
active_category_thresholds[category] = active_threshold
else: # Old structure
if "categories" in st.session_state.thresholds and category in st.session_state.thresholds["categories"]:
if profile_key in st.session_state.thresholds["categories"][category]:
active_category_thresholds[category] = st.session_state.thresholds["categories"][category][profile_key]["threshold"]
else:
active_category_thresholds[category] = active_threshold
else:
active_category_thresholds[category] = active_threshold
# Show informational text for these profiles
st.info(f"The '{threshold_profile}' profile uses pre-optimized thresholds.")
# Show disabled slider for overall threshold (for informational purposes)
st.slider(
"Overall threshold (reference)",
min_value=0.01,
max_value=1.0,
value=float(active_threshold),
step=0.01,
disabled=True
)
elif threshold_profile == "Overall" and st.session_state.model_loaded:
# Use the balanced threshold for Overall profile
if model_type_key is not None: # New structure
if "overall" in st.session_state.thresholds[model_type_key] and "balanced" in st.session_state.thresholds[model_type_key]["overall"]:
active_threshold = st.session_state.thresholds[model_type_key]["overall"]["balanced"]["threshold"]
else: # Old structure
if "overall" in st.session_state.thresholds and "balanced" in st.session_state.thresholds["overall"]:
active_threshold = st.session_state.thresholds["overall"]["balanced"]["threshold"]
# Show slider for adjusting the overall threshold
active_threshold = st.slider(
"Overall threshold",
min_value=0.01,
max_value=1.0,
value=float(active_threshold),
step=0.01
)
elif threshold_profile == "Weighted" and st.session_state.model_loaded:
# Use the balanced threshold as base for Weighted profile
if model_type_key is not None: # New structure
if "overall" in st.session_state.thresholds[model_type_key] and "balanced" in st.session_state.thresholds[model_type_key]["overall"]:
active_threshold = st.session_state.thresholds[model_type_key]["overall"]["balanced"]["threshold"]
else: # Old structure
if "overall" in st.session_state.thresholds and "balanced" in st.session_state.thresholds["overall"]:
active_threshold = st.session_state.thresholds["overall"]["balanced"]["threshold"]
# Show disabled slider for overall threshold (for informational purposes)
st.slider(
"Overall threshold (reference)",
min_value=0.01,
max_value=1.0,
value=float(active_threshold),
step=0.01,
disabled=True
)
st.info("The 'Weighted' profile uses different optimized thresholds for each category.")
# Get weighted thresholds if they exist, otherwise use balanced
if model_type_key is not None: # New structure
if "weighted" in st.session_state.thresholds[model_type_key]:
weighted_thresholds = st.session_state.thresholds[model_type_key]["weighted"]
for category in st.session_state.categories:
if category in weighted_thresholds:
active_category_thresholds[category] = weighted_thresholds[category]
else:
# Fallback to balanced threshold
if "categories" in st.session_state.thresholds[model_type_key] and category in st.session_state.thresholds[model_type_key]["categories"]:
if "balanced" in st.session_state.thresholds[model_type_key]["categories"][category]:
active_category_thresholds[category] = st.session_state.thresholds[model_type_key]["categories"][category]["balanced"]["threshold"]
else:
active_category_thresholds[category] = active_threshold
else:
active_category_thresholds[category] = active_threshold
else: # Old structure
if "weighted" in st.session_state.thresholds:
weighted_thresholds = st.session_state.thresholds["weighted"]
for category in st.session_state.categories:
if category in weighted_thresholds:
active_category_thresholds[category] = weighted_thresholds[category]
else:
# Fallback to balanced threshold
if "categories" in st.session_state.thresholds and category in st.session_state.thresholds["categories"]:
if "balanced" in st.session_state.thresholds["categories"][category]:
active_category_thresholds[category] = st.session_state.thresholds["categories"][category]["balanced"]["threshold"]
else:
active_category_thresholds[category] = active_threshold
else:
active_category_thresholds[category] = active_threshold
elif threshold_profile == "Category-specific" and st.session_state.model_loaded:
# Use the balanced threshold as base for Category-specific
if model_type_key is not None: # New structure
if "overall" in st.session_state.thresholds[model_type_key] and "balanced" in st.session_state.thresholds[model_type_key]["overall"]:
active_threshold = st.session_state.thresholds[model_type_key]["overall"]["balanced"]["threshold"]
else: # Old structure
if "overall" in st.session_state.thresholds and "balanced" in st.session_state.thresholds["overall"]:
active_threshold = st.session_state.thresholds["overall"]["balanced"]["threshold"]
# Show disabled slider for overall threshold (for informational purposes)
st.slider(
"Overall threshold (reference)",
min_value=0.01,
max_value=1.0,
value=float(active_threshold),
step=0.01,
disabled=True
)
st.write("Adjust thresholds for individual categories:")
# Create two columns for better layout of sliders
slider_cols = st.columns(2)
# Initialize with balanced thresholds
for i, category in enumerate(sorted(st.session_state.categories)):
# Get the balanced threshold for this category
category_threshold = None
if model_type_key is not None: # New structure
if "categories" in st.session_state.thresholds[model_type_key] and category in st.session_state.thresholds[model_type_key]["categories"]:
if "balanced" in st.session_state.thresholds[model_type_key]["categories"][category]:
category_threshold = st.session_state.thresholds[model_type_key]["categories"][category]["balanced"]["threshold"]
else:
category_threshold = active_threshold
else:
category_threshold = active_threshold
else: # Old structure
if "categories" in st.session_state.thresholds and category in st.session_state.thresholds["categories"]:
if "balanced" in st.session_state.thresholds["categories"][category]:
category_threshold = st.session_state.thresholds["categories"][category]["balanced"]["threshold"]
else:
category_threshold = active_threshold
else:
category_threshold = active_threshold
# Add slider to appropriate column
col_idx = i % 2 # Alternate between columns
with slider_cols[col_idx]:
active_category_thresholds[category] = st.slider(
f"{category.capitalize()}",
min_value=0.01,
max_value=1.0,
value=float(category_threshold),
step=0.01,
key=f"slider_{category}"
)
# Update session state with the thresholds
if active_threshold is not None:
st.session_state.settings['active_threshold'] = active_threshold
if active_category_thresholds:
st.session_state.settings['active_category_thresholds'] = active_category_thresholds
# Add threshold profile details expander
with st.expander("Threshold Profile Details"):
# Add tabs for the visualizations
if st.session_state.model_loaded:
threshold_tabs = st.tabs(["About Metrics"])
with threshold_tabs[0]:
st.markdown("""
### Understanding Performance Metrics
**F1 Score** is the harmonic mean of precision and recall: `2 * (precision * recall) / (precision + recall)`
**Micro F1** calculates metrics globally by considering each example/prediction pair. This gives more weight to categories with more examples.
**Macro F1** calculates F1 separately for each category and then takes the average. This treats all categories equally regardless of their size.
""")
st.markdown(create_micro_macro_comparison(), unsafe_allow_html=True)
st.markdown("""
### Other Metrics
**Precision** measures how many of the predicted tags are correct: `true_positives / (true_positives + false_positives)`
**Recall** measures how many of the relevant tags are captured: `true_positives / (true_positives + false_negatives)`
### The Precision-Recall Tradeoff
There's an inherent tradeoff between precision and recall:
- Higher threshold → Higher precision, Lower recall
- Lower threshold → Lower precision, Higher recall
The best threshold depends on your specific use case:
- **Prefer Precision**: When false positives are costly (e.g., you want only accurate tags)
- **Prefer Recall**: When false negatives are costly (e.g., you don't want to miss any potentially relevant tags)
- **Balanced**: When both types of errors are equally important
""")
else:
st.info("Load a model to see detailed threshold information.")
# Display options
display_options = st.expander("Display Options", expanded=False)
with display_options:
# Tag display options
col1, col2 = st.columns(2)
with col1:
show_all_tags = st.checkbox("Show all tags (including below threshold)",
value=st.session_state.settings['show_all_tags'])
compact_view = st.checkbox("Compact view (hide progress bars)",
value=st.session_state.settings['compact_view'])
# Add the new checkbox for replacing underscores with spaces
replace_underscores = st.checkbox("Replace underscores with spaces",
value=st.session_state.settings.get('replace_underscores', False))
with col2:
min_confidence = st.slider("Minimum confidence to display", 0.0, 0.5,
st.session_state.settings['min_confidence'], 0.01)
# Update session state with display options
st.session_state.settings['show_all_tags'] = show_all_tags
st.session_state.settings['compact_view'] = compact_view
st.session_state.settings['min_confidence'] = min_confidence
st.session_state.settings['replace_underscores'] = replace_underscores
# Category selection for the "All Tags" section
st.write("Categories to include in 'All Tags' section:")
# Create a multi-column layout for category checkboxes
category_cols = st.columns(3)
selected_categories = {}
# If categories exist in session state, create checkboxes for each
if hasattr(st.session_state, 'categories'):
for i, category in enumerate(sorted(st.session_state.categories)):
col_idx = i % 3 # Distribute across 3 columns
with category_cols[col_idx]:
# Use previously selected value or default to True
default_val = st.session_state.settings['selected_categories'].get(category, True)
selected_categories[category] = st.checkbox(
f"{category.capitalize()}",
value=default_val,
key=f"cat_select_{category}"
)
# Update session state with selected categories
st.session_state.settings['selected_categories'] = selected_categories
if st.session_state.model_loaded:
if st.session_state.model_type == "onnx":
model_type_display = "ONNX Accelerated Model"
elif st.session_state.model_type == "full":
model_type_display = "Full Model"
else:
model_type_display = "Initial-Only Model (Lower VRAM)"
st.info(f"Using: {model_type_display}")
# Run inference button for single image
if image_path and st.button("Run Tagging"):
if not st.session_state.model_loaded:
st.error("Model not loaded. Please check the model settings.")
else:
with st.spinner("Analyzing image..."):
try:
inference_start = time.time()
# Different processing based on model type
if st.session_state.model_type == "onnx":
# Use the appropriate function for ONNX inference
from utils.onnx_processing import process_single_image_onnx
# Run ONNX inference
result = process_single_image_onnx(
image_path=image_path,
model_path=st.session_state.onnx_model_path,
metadata=st.session_state.metadata,
threshold_profile=threshold_profile,
active_threshold=active_threshold,
active_category_thresholds=active_category_thresholds,
min_confidence=min_confidence
)
else:
# Run standard PyTorch inference
result = process_image(
image_path=image_path,
model=st.session_state.model,
thresholds=st.session_state.thresholds,
metadata=st.session_state.metadata,
threshold_profile=threshold_profile,
active_threshold=active_threshold,
active_category_thresholds=active_category_thresholds,
min_confidence=min_confidence
)
inference_time = time.time() - inference_start
if result['success']:
# Store results in session state
st.session_state.all_probs = result['all_probs']
st.session_state.tags = result['tags']
st.session_state.all_tags = result['all_tags']
st.success(f"Analysis completed in {inference_time:.2f} seconds")
else:
st.error(f"Inference failed: {result.get('error', 'Unknown error')}")
except Exception as e:
st.error(f"Inference error: {str(e)}")
st.code(traceback.format_exc())
# Display the predictions if available
if image_path and hasattr(st.session_state, 'all_probs'):
st.header("Predictions")
# Apply current thresholds to stored probabilities
filtered_tags, current_all_tags = apply_thresholds(
st.session_state.all_probs,
threshold_profile,
active_threshold,
active_category_thresholds,
min_confidence,
st.session_state.settings['selected_categories']
)
# Store the updated results back to session state
st.session_state.tags = filtered_tags
st.session_state.all_tags = current_all_tags
# Create an empty list to collect all tags that pass the thresholds
# We'll rebuild this list as we process each category
all_tags = []
for category in sorted(st.session_state.all_probs.keys()):
# Get all tags for this category and the filtered ones
all_tags_in_category = st.session_state.all_probs.get(category, [])
filtered_tags_in_category = filtered_tags.get(category, [])
# Only show categories with tags
if all_tags_in_category:
# Get the appropriate threshold for this category
if threshold_profile in ["Overall", "Weighted"]:
threshold = active_threshold
else:
threshold = active_category_thresholds.get(category, active_threshold)
# Create expander with count information
expander_label = f"{category.capitalize()} ({len(filtered_tags_in_category)} tags)"
with st.expander(expander_label, expanded=True):
# Add threshold control specific to this category
threshold_row = st.columns([1, 2])
with threshold_row[0]:
# Display the current threshold
pass
with threshold_row[1]:
# Always show category slider regardless of threshold profile
cat_slider_key = f"cat_threshold_{category}"
# For the initial value, respect the threshold profile
if threshold_profile in ["Overall", "Weighted"]:
# In these modes, start with the global threshold value
current_cat_threshold = active_threshold
else:
# In other modes, use the category-specific threshold
current_cat_threshold = active_category_thresholds.get(category, active_threshold)
# Add a slider for this specific category
new_threshold = st.slider(
f"Adjust {category.capitalize()} threshold:",
min_value=0.01,
max_value=1.0,
value=float(current_cat_threshold),
step=0.01,
key=cat_slider_key,
disabled=(threshold_profile in ["Overall", "Weighted"]) # Disable in Overall/Weighted modes
)
# If in Overall/Weighted mode, add an explanation
if threshold_profile in ["Overall", "Weighted"]:
st.info(f"Using global {threshold_profile.lower()} threshold. Switch to Category-specific mode to adjust individual categories.")
else:
# In other modes, update the category threshold
active_category_thresholds[category] = new_threshold
threshold = new_threshold
# Update the active_category_thresholds with the new value
active_category_thresholds[category] = new_threshold
# Very important: Update the threshold variable for this iteration of the loop
# This ensures the new threshold is used in this category's display
threshold = new_threshold
# Determine which tags to display based on show_all_tags setting
if show_all_tags:
tags_to_display = all_tags_in_category
else:
# Refilter the tags based on the updated threshold
# This ensures the display immediately reflects threshold changes
tags_to_display = [(tag, prob) for tag, prob in all_tags_in_category if prob >= threshold]
filtered_tags[category] = tags_to_display
# Add per-category limit controls
limit_col1, limit_col2 = st.columns([1, 2])
with limit_col1:
# Generate a unique key for each category's limit checkbox
limit_key = f"limit_{category}_tags"
limit_tags_for_category = st.checkbox("Limit tags", value=False, key=limit_key)
with limit_col2:
# Generate a unique key for each category's tag count slider
slider_key = f"top_n_{category}_tags"
# Handle the slider values carefully to avoid Streamlit errors
tag_count = len(tags_to_display)
# Skip slider if there are no tags to display
if tag_count > 0:
# Set sensible min and max values that avoid errors
min_value = 0
# Make sure max_value is at least 1 (Streamlit requires min < max)
max_value = max(1, min(999, tag_count))
# Default value should be between min and max
default_value = min(max_value, 5) # Default to 5 tags if available
top_n_tags_for_category = st.slider(
"Show top",
min_value=min_value,
max_value=max_value,
value=default_value,
step=1,
disabled=not limit_tags_for_category,
key=slider_key
)
else:
# If no tags, just set a default value without showing the slider
top_n_tags_for_category = 5
st.write("No tags to display")
st.markdown("---") # Add separator line after controls
if not tags_to_display:
st.info(f"No tags above {min_confidence:.2f} confidence threshold")
continue
# Apply the per-category limit if enabled
original_count = len(tags_to_display)
if limit_tags_for_category and tags_to_display:
limited_tags_to_display = tags_to_display[:top_n_tags_for_category]
display_count = len(limited_tags_to_display)
else:
limited_tags_to_display = tags_to_display
display_count = original_count
# Display tags based on view mode
if compact_view:
# Compact view - show tags in a comma-separated list with %
tag_list = []
replace_underscores = st.session_state.settings.get('replace_underscores', False)
for tag, prob in limited_tags_to_display:
# Format: tag (percent%)
percentage = int(prob * 100)
# Apply underscore replacement for display if enabled
display_tag = tag.replace('_', ' ') if replace_underscores else tag
tag_list.append(f"{display_tag} ({percentage}%)")
# Add tag to all_tags list if it passes threshold and category is selected
# Note: always use original tag (with underscores) for all_tags collection
if prob >= threshold and selected_categories.get(category, True):
all_tags.append(tag)
# Join with commas and display
st.markdown(", ".join(tag_list))
else:
# Expanded view with progress bars
for tag, prob in limited_tags_to_display:
# Get the replacement setting
replace_underscores = st.session_state.settings.get('replace_underscores', False)
# Apply underscore replacement for display if enabled
display_tag = tag.replace('_', ' ') if replace_underscores else tag
# Add tag to all_tags list if it passes threshold and category is selected
if prob >= threshold and selected_categories.get(category, True):
all_tags.append(tag) # Add original tag to all_tags
tag_display = f"**{display_tag}**" # Bold for tags above threshold
else:
tag_display = display_tag # Regular for tags below threshold
# Display tag with progress bar
st.write(tag_display)
st.markdown(display_progress_bar(prob), unsafe_allow_html=True)
# If we're limiting tags and there are more available, show a message
if limit_tags_for_category and original_count > display_count:
st.caption(f"Showing top {display_count} of {original_count} qualifying tags.")
# Show summary at bottom with the truly updated all_tags list
st.markdown("---")
st.subheader(f"All Tags ({len(all_tags)} total)")
if all_tags:
# Check if underscore replacement is enabled
replace_underscores = st.session_state.settings.get('replace_underscores', False)
if replace_underscores:
# Create a new list with underscores replaced by spaces for display only
display_tags = [tag.replace('_', ' ') for tag in all_tags]
st.write(", ".join(display_tags))
else:
# Display original tags
st.write(", ".join(all_tags))
else:
st.info("No tags detected above the threshold.")
# Add Save Tags section
st.markdown("---")
st.subheader("Save Tags")
# Create column for save options
save_col = st.columns(1)[0]
with save_col:
# Option to save to custom location
if 'custom_folders' not in st.session_state:
# Initialize with default save locations
st.session_state.custom_folders = get_default_save_locations()
# Display folder selection dropdown
selected_folder = st.selectbox(
"Select save location:",
options=st.session_state.custom_folders,
format_func=lambda x: os.path.basename(x) if os.path.basename(x) else x
)
# Allow adding a new folder path
new_folder = st.text_input("Or enter a new folder path:")
if st.button("Add Folder", key="add_folder") and new_folder:
if os.path.isdir(new_folder):
if new_folder not in st.session_state.custom_folders:
st.session_state.custom_folders.append(new_folder)
st.success(f"Added folder: {new_folder}")
st.rerun()
else:
st.info("This folder is already in the list.")
else:
try:
# Try to create the folder if it doesn't exist
os.makedirs(new_folder, exist_ok=True)
st.session_state.custom_folders.append(new_folder)
st.success(f"Created and added folder: {new_folder}")
st.rerun()
except Exception as e:
st.error(f"Could not create folder: {str(e)}")
# Save to selected location button
if st.button("💾 Save to Selected Location"):
try:
# Get the original filename if it exists
original_filename = st.session_state.original_filename if hasattr(st.session_state, 'original_filename') else None
# Save tags to file in selected location
saved_path = save_tags_to_file(
image_path=image_path,
all_tags=all_tags,
original_filename=original_filename,
custom_dir=selected_folder,
overwrite=True
)
st.success(f"Tags saved to: {os.path.basename(saved_path)}")
st.info(f"Full path: {saved_path}")
# Show preview of saved file
with st.expander("File Contents", expanded=True):
with open(saved_path, 'r', encoding='utf-8') as f:
content = f.read()
st.code(content, language='text')
except Exception as e:
st.error(f"Error saving tags: {str(e)}")
st.code(traceback.format_exc())
if __name__ == "__main__":
image_tagger_app() |