sycod
commited on
Commit
·
38dbb38
1
Parent(s):
a03ee84
first train ok, oversampling begun
Browse files- .gitignore +1 -0
- EDA.ipynb +562 -266
- README.md +3 -2
- config.yaml +15 -8
- src/load_data.py +16 -2
.gitignore
CHANGED
|
@@ -13,5 +13,6 @@ eval-results/
|
|
| 13 |
eval-queue-bk/
|
| 14 |
eval-results-bk/
|
| 15 |
logs/
|
|
|
|
| 16 |
pyro-sdis/
|
| 17 |
venv/
|
|
|
|
| 13 |
eval-queue-bk/
|
| 14 |
eval-results-bk/
|
| 15 |
logs/
|
| 16 |
+
models/
|
| 17 |
pyro-sdis/
|
| 18 |
venv/
|
EDA.ipynb
CHANGED
|
@@ -5,15 +5,20 @@
|
|
| 5 |
"metadata": {},
|
| 6 |
"source": [
|
| 7 |
"**Table of contents**<a id='toc0_'></a> \n",
|
| 8 |
-
"- [
|
| 9 |
"- [Load and import](#toc2_) \n",
|
| 10 |
" - [Import libraries](#toc2_1_) \n",
|
| 11 |
" - [Load configuration](#toc2_2_) \n",
|
| 12 |
-
" - [Load data](#toc2_3_) \n",
|
| 13 |
"- [EDA](#toc3_) \n",
|
| 14 |
-
" - [
|
| 15 |
-
" - [🚧
|
| 16 |
-
" - [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
"\n",
|
| 18 |
"<!-- vscode-jupyter-toc-config\n",
|
| 19 |
"\tnumbering=false\n",
|
|
@@ -29,7 +34,7 @@
|
|
| 29 |
"cell_type": "markdown",
|
| 30 |
"metadata": {},
|
| 31 |
"source": [
|
| 32 |
-
"# <a id='toc1_'></a>[
|
| 33 |
"\n",
|
| 34 |
"https://huggingface.co/datasets/pyronear/pyro-sdis\n",
|
| 35 |
"\n",
|
|
@@ -61,25 +66,25 @@
|
|
| 61 |
"text": [
|
| 62 |
"/Users/julmat/Documents/hugging_face/frugal_cviz/.venv/lib/python3.12/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
|
| 63 |
" from .autonotebook import tqdm as notebook_tqdm\n",
|
| 64 |
-
"[codecarbon WARNING @
|
| 65 |
-
"[codecarbon INFO @
|
| 66 |
-
"[codecarbon INFO @
|
| 67 |
-
"[codecarbon WARNING @
|
| 68 |
" Mac OS and ARM processor detected: Please enable PowerMetrics sudo to measure CPU\n",
|
| 69 |
"\n",
|
| 70 |
-
"[codecarbon INFO @
|
| 71 |
-
"[codecarbon INFO @
|
| 72 |
-
"[codecarbon INFO @
|
| 73 |
-
"[codecarbon INFO @
|
| 74 |
-
"[codecarbon INFO @
|
| 75 |
-
"[codecarbon INFO @
|
| 76 |
-
"[codecarbon INFO @
|
| 77 |
-
"[codecarbon INFO @
|
| 78 |
-
"[codecarbon INFO @
|
| 79 |
-
"[codecarbon INFO @
|
| 80 |
-
"[codecarbon INFO @
|
| 81 |
-
"[codecarbon INFO @
|
| 82 |
-
"[codecarbon INFO @
|
| 83 |
]
|
| 84 |
}
|
| 85 |
],
|
|
@@ -88,6 +93,7 @@
|
|
| 88 |
"from fastapi import APIRouter\n",
|
| 89 |
"import logging\n",
|
| 90 |
"import matplotlib.pyplot as plt\n",
|
|
|
|
| 91 |
"from PIL import Image, ImageOps, ImageEnhance, ImageFilter\n",
|
| 92 |
"import plotly.express as px\n",
|
| 93 |
"import random\n",
|
|
@@ -96,9 +102,10 @@
|
|
| 96 |
"# ML\n",
|
| 97 |
"from keras import Model\n",
|
| 98 |
"from keras.applications import EfficientNetB0\n",
|
| 99 |
-
"from keras.utils import image_dataset_from_directory\n",
|
| 100 |
"from keras.layers import Flatten, Dense\n",
|
|
|
|
| 101 |
"from keras.optimizers import AdamW\n",
|
|
|
|
| 102 |
"import numpy as np\n",
|
| 103 |
"import pandas as pd\n",
|
| 104 |
"from scipy.ndimage import median_filter\n",
|
|
@@ -113,6 +120,7 @@
|
|
| 113 |
"from tasks.utils.emissions import tracker, clean_emissions_data, get_space_info\n",
|
| 114 |
"from tasks.image import parse_boxes, compute_iou, compute_max_iou\n",
|
| 115 |
"\n",
|
|
|
|
| 116 |
"# Logging configuration (see all outputs, even DEBUG or INFO)\n",
|
| 117 |
"logger = logging.getLogger()\n",
|
| 118 |
"logger.setLevel(logging.INFO)"
|
|
@@ -134,11 +142,17 @@
|
|
| 134 |
"# local config\n",
|
| 135 |
"with open(\"config.yaml\", \"r\") as f:\n",
|
| 136 |
" cfg = yaml.safe_load(f)\n",
|
|
|
|
| 137 |
"OUTPUT_DIR = cfg[\"data_root_dir\"]\n",
|
| 138 |
"DB_INFO_URI = cfg[\"db_info_uri\"]\n",
|
| 139 |
"REPO_ID = cfg[\"repo_id\"]\n",
|
| 140 |
"SPLIT_SIZE = cfg[\"split_size\"]\n",
|
| 141 |
"RDM_SEED = cfg[\"rdm_seed\"]\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 142 |
"\n",
|
| 143 |
"request = ImageEvaluationRequest()"
|
| 144 |
]
|
|
@@ -771,7 +785,7 @@
|
|
| 771 |
"cell_type": "markdown",
|
| 772 |
"metadata": {},
|
| 773 |
"source": [
|
| 774 |
-
"## Classes distribution"
|
| 775 |
]
|
| 776 |
},
|
| 777 |
{
|
|
@@ -1742,20 +1756,20 @@
|
|
| 1742 |
"cell_type": "markdown",
|
| 1743 |
"metadata": {},
|
| 1744 |
"source": [
|
| 1745 |
-
"## 🚧 Constats
|
| 1746 |
"\n",
|
| 1747 |
"- 1 seul définition : 1280x720\n",
|
| 1748 |
"- 1 seul format : JPG\n",
|
| 1749 |
"- 1 seul mode : RGB\n",
|
| 1750 |
"\n",
|
| 1751 |
-
"- voir
|
| 1752 |
]
|
| 1753 |
},
|
| 1754 |
{
|
| 1755 |
"cell_type": "markdown",
|
| 1756 |
"metadata": {},
|
| 1757 |
"source": [
|
| 1758 |
-
"## <a id='
|
| 1759 |
]
|
| 1760 |
},
|
| 1761 |
{
|
|
@@ -1906,7 +1920,7 @@
|
|
| 1906 |
"cell_type": "markdown",
|
| 1907 |
"metadata": {},
|
| 1908 |
"source": [
|
| 1909 |
-
"# Data preprocessing"
|
| 1910 |
]
|
| 1911 |
},
|
| 1912 |
{
|
|
@@ -1941,28 +1955,7 @@
|
|
| 1941 |
"name": "stderr",
|
| 1942 |
"output_type": "stream",
|
| 1943 |
"text": [
|
| 1944 |
-
"
|
| 1945 |
-
"\n",
|
| 1946 |
-
"\n",
|
| 1947 |
-
"A value is trying to be set on a copy of a slice from a DataFrame.\n",
|
| 1948 |
-
"Try using .loc[row_indexer,col_indexer] = value instead\n",
|
| 1949 |
-
"\n",
|
| 1950 |
-
"See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
|
| 1951 |
-
"\n",
|
| 1952 |
-
"/Users/julmat/Documents/hugging_face/frugal_cviz/src/load_data.py:195: SettingWithCopyWarning:\n",
|
| 1953 |
-
"\n",
|
| 1954 |
-
"\n",
|
| 1955 |
-
"A value is trying to be set on a copy of a slice from a DataFrame\n",
|
| 1956 |
-
"\n",
|
| 1957 |
-
"See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
|
| 1958 |
-
"\n",
|
| 1959 |
-
"/Users/julmat/Documents/hugging_face/frugal_cviz/src/load_data.py:196: SettingWithCopyWarning:\n",
|
| 1960 |
-
"\n",
|
| 1961 |
-
"\n",
|
| 1962 |
-
"A value is trying to be set on a copy of a slice from a DataFrame\n",
|
| 1963 |
-
"\n",
|
| 1964 |
-
"See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
|
| 1965 |
-
"\n"
|
| 1966 |
]
|
| 1967 |
}
|
| 1968 |
],
|
|
@@ -1974,7 +1967,251 @@
|
|
| 1974 |
"cell_type": "markdown",
|
| 1975 |
"metadata": {},
|
| 1976 |
"source": [
|
| 1977 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1978 |
]
|
| 1979 |
},
|
| 1980 |
{
|
|
@@ -2002,83 +2239,122 @@
|
|
| 2002 |
"cell_type": "markdown",
|
| 2003 |
"metadata": {},
|
| 2004 |
"source": [
|
| 2005 |
-
"# Model Training"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2006 |
]
|
| 2007 |
},
|
| 2008 |
{
|
| 2009 |
"cell_type": "code",
|
| 2010 |
-
"execution_count":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2011 |
"metadata": {},
|
| 2012 |
"outputs": [
|
| 2013 |
{
|
| 2014 |
"name": "stdout",
|
| 2015 |
"output_type": "stream",
|
| 2016 |
"text": [
|
| 2017 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2018 |
]
|
| 2019 |
}
|
| 2020 |
],
|
| 2021 |
"source": [
|
| 2022 |
-
"
|
| 2023 |
-
"\n",
|
| 2024 |
-
"
|
| 2025 |
-
"
|
| 2026 |
-
"from keras.layers import Flatten, Dense\n",
|
| 2027 |
-
"from keras.optimizers import AdamW\n",
|
| 2028 |
-
"from keras.utils import image_dataset_from_directory\n",
|
| 2029 |
-
"from sklearn.model_selection import train_test_split\n",
|
| 2030 |
-
"import tensorflow as tf\n",
|
| 2031 |
-
"from src.models import eval_pretrained_model, ConditionalAugmentation\n",
|
| 2032 |
-
"\n",
|
| 2033 |
-
"\n",
|
| 2034 |
-
"# *****************************************************************************\n",
|
| 2035 |
-
"# CONFIGURATION\n",
|
| 2036 |
-
"# *****************************************************************************\n",
|
| 2037 |
-
"# load config\n",
|
| 2038 |
-
"MODEL_PREV = cfg[\"previous\"][\"model\"]\n",
|
| 2039 |
-
"CLASSES_10 = cfg[\"previous\"][\"classes_10\"]\n",
|
| 2040 |
-
"TRAIN_DIR_10 = os.path.join(DATA_DIR, cfg[\"previous\"][\"train_dir\"]) + \"_10\"\n",
|
| 2041 |
-
"TEST_DIR_10 = os.path.join(DATA_DIR, cfg[\"previous\"][\"test_dir\"]) + \"_10\"\n",
|
| 2042 |
-
"CHKPT_DIR = os.path.join(DATA_DIR, cfg[\"previous\"][\"checkpoint_dir\"])\n",
|
| 2043 |
-
"LOG_DIR = cfg[\"previous\"][\"log_dir\"]\n",
|
| 2044 |
"\n",
|
| 2045 |
-
"#
|
| 2046 |
-
"
|
| 2047 |
-
"
|
| 2048 |
-
"
|
| 2049 |
-
"
|
| 2050 |
-
"
|
| 2051 |
-
"
|
| 2052 |
-
"
|
| 2053 |
-
"\n",
|
| 2054 |
-
"#
|
| 2055 |
-
"# DATA PREPARATION\n",
|
| 2056 |
-
"# *****************************************************************************\n",
|
| 2057 |
-
"y = img_df[\"class_label\"]\n",
|
| 2058 |
-
"X = img_df[\"img_uri\"]\n",
|
| 2059 |
-
"# train / test split\n",
|
| 2060 |
-
"X_train, X_test, y_train, y_test = train_test_split(\n",
|
| 2061 |
-
" X, y, test_size=0.1, stratify=y, random_state=42\n",
|
| 2062 |
")\n",
|
| 2063 |
-
"# create samples directories\n",
|
| 2064 |
-
"if not os.path.exists(TRAIN_DIR_10):\n",
|
| 2065 |
-
" os.makedirs(TRAIN_DIR_10)\n",
|
| 2066 |
-
"if not os.path.exists(TEST_DIR_10):\n",
|
| 2067 |
-
" os.makedirs(TEST_DIR_10)\n",
|
| 2068 |
-
"\n",
|
| 2069 |
-
"print(\"Train set: \")\n",
|
| 2070 |
-
"copy_images_prev(X_train, TRAIN_DIR_10)\n",
|
| 2071 |
-
"print(\"Test set: \")\n",
|
| 2072 |
-
"copy_images_prev(X_test, TEST_DIR_10)\n",
|
| 2073 |
"\n",
|
| 2074 |
-
"print(\"
|
| 2075 |
-
"
|
| 2076 |
-
"
|
| 2077 |
" labels=\"inferred\", # class names upon folders structure\n",
|
| 2078 |
" label_mode=\"int\", # integer encoding\n",
|
| 2079 |
-
"
|
| 2080 |
-
" subset=\"both\", # returns both train and val datasets\n",
|
| 2081 |
-
" shuffle=True, # shuffles images\n",
|
| 2082 |
" seed=42, # random seed\n",
|
| 2083 |
" image_size=input_size, # automatic resizing\n",
|
| 2084 |
" batch_size=batch_size, # tensor shape[0]\n",
|
|
@@ -2086,59 +2362,13 @@
|
|
| 2086 |
"\n",
|
| 2087 |
"print(\"\\nTest dataset:\")\n",
|
| 2088 |
"test_ds = image_dataset_from_directory(\n",
|
| 2089 |
-
"
|
| 2090 |
" labels=\"inferred\", # class names upon folders structure\n",
|
| 2091 |
" label_mode=\"int\", # integer encoding\n",
|
| 2092 |
-
" shuffle=False, # do not
|
| 2093 |
" seed=42, # random seed\n",
|
| 2094 |
" image_size=input_size, # automatic resizing\n",
|
| 2095 |
" batch_size=batch_size, # tensor shape[0]\n",
|
| 2096 |
-
")\n",
|
| 2097 |
-
"\n",
|
| 2098 |
-
"# *****************************************************************************\n",
|
| 2099 |
-
"# MODEL\n",
|
| 2100 |
-
"# *****************************************************************************\n",
|
| 2101 |
-
"# load pre-trained model without top layers\n",
|
| 2102 |
-
"model = EfficientNetB0(\n",
|
| 2103 |
-
" weights=\"imagenet\", # pre-trained weights\n",
|
| 2104 |
-
" include_top=False, # no dense layer\n",
|
| 2105 |
-
" input_shape=(input_size[0], input_size[1], 3), # input shape\n",
|
| 2106 |
-
")\n",
|
| 2107 |
-
"# create explicit input layer\n",
|
| 2108 |
-
"inputs = tf.keras.Input(shape=(input_size[0], input_size[1], 3))\n",
|
| 2109 |
-
"# add data augmentation\n",
|
| 2110 |
-
"augmented = ConditionalAugmentation(rate=0.4)(inputs)\n",
|
| 2111 |
-
"x = model(augmented)\n",
|
| 2112 |
-
"\n",
|
| 2113 |
-
"# flatten output\n",
|
| 2114 |
-
"x = Flatten()(x)\n",
|
| 2115 |
-
"# for feature extraction only\n",
|
| 2116 |
-
"for layer in model.layers:\n",
|
| 2117 |
-
" layer.trainable = False\n",
|
| 2118 |
-
"# new FC layer for 3 classes classification\n",
|
| 2119 |
-
"predictions = Dense(10, activation=\"softmax\")(x)\n",
|
| 2120 |
-
"# define new model\n",
|
| 2121 |
-
"effnetB0_10 = Model(inputs=inputs, outputs=predictions)\n",
|
| 2122 |
-
"# display model summary\n",
|
| 2123 |
-
"effnetB0_10.summary()\n",
|
| 2124 |
-
"\n",
|
| 2125 |
-
"# *****************************************************************************\n",
|
| 2126 |
-
"# TRAINING\n",
|
| 2127 |
-
"# *****************************************************************************\n",
|
| 2128 |
-
"model_trained, history = eval_pretrained_model(\n",
|
| 2129 |
-
" model=effnetB0_10,\n",
|
| 2130 |
-
" train_ds=train_ds,\n",
|
| 2131 |
-
" val_ds=val_ds,\n",
|
| 2132 |
-
" test_ds=test_ds,\n",
|
| 2133 |
-
" LOG_DIR=LOG_DIR,\n",
|
| 2134 |
-
" CHKPT_DIR=CHKPT_DIR,\n",
|
| 2135 |
-
" model_name=model_name,\n",
|
| 2136 |
-
" input_size=input_size,\n",
|
| 2137 |
-
" batch_size=batch_size,\n",
|
| 2138 |
-
" n_epochs=n_epochs,\n",
|
| 2139 |
-
" optimizer=optimizer,\n",
|
| 2140 |
-
" loss=loss,\n",
|
| 2141 |
-
" metrics=metrics,\n",
|
| 2142 |
")"
|
| 2143 |
]
|
| 2144 |
},
|
|
@@ -2146,51 +2376,32 @@
|
|
| 2146 |
"cell_type": "markdown",
|
| 2147 |
"metadata": {},
|
| 2148 |
"source": [
|
| 2149 |
-
"
|
| 2150 |
-
]
|
| 2151 |
-
},
|
| 2152 |
-
{
|
| 2153 |
-
"cell_type": "code",
|
| 2154 |
-
"execution_count": 21,
|
| 2155 |
-
"metadata": {},
|
| 2156 |
-
"outputs": [],
|
| 2157 |
-
"source": [
|
| 2158 |
-
"input_size = (224, 224)\n",
|
| 2159 |
-
"batch_size = 48\n",
|
| 2160 |
-
"n_epochs = 100\n",
|
| 2161 |
-
"optimizer = AdamW(learning_rate=0.0002, weight_decay=0.05)\n",
|
| 2162 |
-
"loss = \"sparse_categorical_crossentropy\"\n",
|
| 2163 |
-
"metrics = [\"accuracy\"]"
|
| 2164 |
-
]
|
| 2165 |
-
},
|
| 2166 |
-
{
|
| 2167 |
-
"cell_type": "markdown",
|
| 2168 |
-
"metadata": {},
|
| 2169 |
-
"source": [
|
| 2170 |
-
"Train EfficientNet"
|
| 2171 |
]
|
| 2172 |
},
|
| 2173 |
{
|
| 2174 |
"cell_type": "code",
|
| 2175 |
-
"execution_count":
|
| 2176 |
"metadata": {},
|
| 2177 |
"outputs": [
|
| 2178 |
{
|
| 2179 |
"name": "stderr",
|
| 2180 |
"output_type": "stream",
|
| 2181 |
"text": [
|
| 2182 |
-
"/Users/julmat/Documents/hugging_face/frugal_cviz/.venv/lib/python3.12/site-packages/keras/src/layers/layer.py:393: UserWarning
|
| 2183 |
-
"
|
|
|
|
|
|
|
| 2184 |
]
|
| 2185 |
},
|
| 2186 |
{
|
| 2187 |
"data": {
|
| 2188 |
"text/html": [
|
| 2189 |
-
"<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"
|
| 2190 |
"</pre>\n"
|
| 2191 |
],
|
| 2192 |
"text/plain": [
|
| 2193 |
-
"\u001b[1mModel: \"
|
| 2194 |
]
|
| 2195 |
},
|
| 2196 |
"metadata": {},
|
|
@@ -2202,16 +2413,16 @@
|
|
| 2202 |
"<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
|
| 2203 |
"┃<span style=\"font-weight: bold\"> Layer (type) </span>┃<span style=\"font-weight: bold\"> Output Shape </span>┃<span style=\"font-weight: bold\"> Param # </span>┃\n",
|
| 2204 |
"┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
|
| 2205 |
-
"│
|
| 2206 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2207 |
-
"│
|
| 2208 |
"│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">ConditionalAugmentation</span>) │ │ │\n",
|
| 2209 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2210 |
"│ efficientnetb0 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Functional</span>) │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">7</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">7</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">1280</span>) │ <span style=\"color: #00af00; text-decoration-color: #00af00\">4,049,571</span> │\n",
|
| 2211 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2212 |
-
"│
|
| 2213 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2214 |
-
"│
|
| 2215 |
"└─────────────────────────────────┴────────────────────────┴───────────────┘\n",
|
| 2216 |
"</pre>\n"
|
| 2217 |
],
|
|
@@ -2219,16 +2430,16 @@
|
|
| 2219 |
"┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
|
| 2220 |
"┃\u001b[1m \u001b[0m\u001b[1mLayer (type) \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
|
| 2221 |
"┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
|
| 2222 |
-
"│
|
| 2223 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2224 |
-
"│
|
| 2225 |
"│ (\u001b[38;5;33mConditionalAugmentation\u001b[0m) │ │ │\n",
|
| 2226 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2227 |
"│ efficientnetb0 (\u001b[38;5;33mFunctional\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m7\u001b[0m, \u001b[38;5;34m7\u001b[0m, \u001b[38;5;34m1280\u001b[0m) │ \u001b[38;5;34m4,049,571\u001b[0m │\n",
|
| 2228 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2229 |
-
"│
|
| 2230 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2231 |
-
"│
|
| 2232 |
"└─────────────────────────────────┴────────────────────────┴───────────────┘\n"
|
| 2233 |
]
|
| 2234 |
},
|
|
@@ -2238,11 +2449,11 @@
|
|
| 2238 |
{
|
| 2239 |
"data": {
|
| 2240 |
"text/html": [
|
| 2241 |
-
"<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">4,
|
| 2242 |
"</pre>\n"
|
| 2243 |
],
|
| 2244 |
"text/plain": [
|
| 2245 |
-
"\u001b[1m Total params: \u001b[0m\u001b[38;5;34m4,
|
| 2246 |
]
|
| 2247 |
},
|
| 2248 |
"metadata": {},
|
|
@@ -2251,11 +2462,11 @@
|
|
| 2251 |
{
|
| 2252 |
"data": {
|
| 2253 |
"text/html": [
|
| 2254 |
-
"<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">
|
| 2255 |
"</pre>\n"
|
| 2256 |
],
|
| 2257 |
"text/plain": [
|
| 2258 |
-
"\u001b[1m Trainable params: \u001b[0m\u001b[38;5;
|
| 2259 |
]
|
| 2260 |
},
|
| 2261 |
"metadata": {},
|
|
@@ -2276,30 +2487,28 @@
|
|
| 2276 |
}
|
| 2277 |
],
|
| 2278 |
"source": [
|
| 2279 |
-
"
|
| 2280 |
-
"\n",
|
| 2281 |
-
"# load pre-trained model without top layers\n",
|
| 2282 |
"model = EfficientNetB0(\n",
|
| 2283 |
" weights=\"imagenet\", # pre-trained weights\n",
|
| 2284 |
" include_top=False, # no dense layer\n",
|
| 2285 |
" input_shape=(input_size[0], input_size[1], 3), # input shape\n",
|
| 2286 |
")\n",
|
| 2287 |
-
"#
|
| 2288 |
"inputs = tf.keras.Input(shape=(input_size[0], input_size[1], 3))\n",
|
| 2289 |
"# add data augmentation\n",
|
| 2290 |
"augmented = ConditionalAugmentation(rate=0.4)(inputs)\n",
|
| 2291 |
"x = model(augmented)\n",
|
| 2292 |
"\n",
|
| 2293 |
-
"#
|
| 2294 |
"x = Flatten()(x)\n",
|
| 2295 |
-
"#
|
| 2296 |
"for layer in model.layers:\n",
|
| 2297 |
" layer.trainable = False\n",
|
| 2298 |
-
"#
|
| 2299 |
-
"predictions = Dense(
|
| 2300 |
-
"#
|
| 2301 |
"effnetB0 = Model(inputs=inputs, outputs=predictions)\n",
|
| 2302 |
-
"#
|
| 2303 |
"effnetB0.summary()"
|
| 2304 |
]
|
| 2305 |
},
|
|
@@ -2307,71 +2516,165 @@
|
|
| 2307 |
"cell_type": "markdown",
|
| 2308 |
"metadata": {},
|
| 2309 |
"source": [
|
| 2310 |
-
"
|
| 2311 |
-
]
|
| 2312 |
-
},
|
| 2313 |
-
{
|
| 2314 |
-
"cell_type": "code",
|
| 2315 |
-
"execution_count": null,
|
| 2316 |
-
"metadata": {},
|
| 2317 |
-
"outputs": [],
|
| 2318 |
-
"source": [
|
| 2319 |
-
"train_ds = image_dataset_from_directory(\n",
|
| 2320 |
-
" TRAIN_DIR,\n",
|
| 2321 |
-
" labels=\"inferred\", # class names upon folders structure\n",
|
| 2322 |
-
" label_mode=\"int\", # integer encoding\n",
|
| 2323 |
-
" # validation_split=0.2, # train / val split\n",
|
| 2324 |
-
" # subset=\"both\", # returns both train and val datasets\n",
|
| 2325 |
-
" shuffle=True, # shuffles images\n",
|
| 2326 |
-
" seed=42, # random seed\n",
|
| 2327 |
-
" image_size=input_size, # automatic resizing\n",
|
| 2328 |
-
" batch_size=batch_size, # tensor shape[0]\n",
|
| 2329 |
-
")"
|
| 2330 |
]
|
| 2331 |
},
|
| 2332 |
{
|
| 2333 |
-
"cell_type": "
|
| 2334 |
-
"execution_count": null,
|
| 2335 |
"metadata": {},
|
| 2336 |
-
"outputs": [],
|
| 2337 |
"source": [
|
| 2338 |
-
"
|
| 2339 |
-
"# TEST_DIR,\n",
|
| 2340 |
-
"# labels=\"inferred\", # class names upon folders structure\n",
|
| 2341 |
-
"# label_mode=\"int\", # integer encoding\n",
|
| 2342 |
-
"# shuffle=False, # do not shuffles images\n",
|
| 2343 |
-
"# seed=42, # random seed\n",
|
| 2344 |
-
"# image_size=input_size, # automatic resizing\n",
|
| 2345 |
-
"# batch_size=batch_size, # tensor shape[0]\n",
|
| 2346 |
-
"# )"
|
| 2347 |
]
|
| 2348 |
},
|
| 2349 |
{
|
| 2350 |
"cell_type": "code",
|
| 2351 |
-
"execution_count":
|
| 2352 |
"metadata": {},
|
| 2353 |
-
"outputs": [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2354 |
"source": [
|
| 2355 |
-
"
|
| 2356 |
-
"
|
| 2357 |
-
"
|
| 2358 |
-
"
|
| 2359 |
-
"
|
| 2360 |
-
"
|
| 2361 |
-
"
|
| 2362 |
-
"
|
| 2363 |
-
"
|
| 2364 |
-
"
|
| 2365 |
-
"
|
| 2366 |
-
"
|
| 2367 |
-
"
|
| 2368 |
-
"
|
| 2369 |
-
"
|
| 2370 |
]
|
| 2371 |
},
|
| 2372 |
{
|
| 2373 |
"cell_type": "code",
|
| 2374 |
-
"execution_count":
|
| 2375 |
"metadata": {},
|
| 2376 |
"outputs": [
|
| 2377 |
{
|
|
@@ -2381,7 +2684,7 @@
|
|
| 2381 |
"traceback": [
|
| 2382 |
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
|
| 2383 |
"\u001b[0;31mNameError\u001b[0m Traceback (most recent call last)",
|
| 2384 |
-
"Cell \u001b[0;32mIn[
|
| 2385 |
"\u001b[0;31mNameError\u001b[0m: name 'stop' is not defined"
|
| 2386 |
]
|
| 2387 |
}
|
|
@@ -2434,14 +2737,7 @@
|
|
| 2434 |
"cell_type": "markdown",
|
| 2435 |
"metadata": {},
|
| 2436 |
"source": [
|
| 2437 |
-
"
|
| 2438 |
-
]
|
| 2439 |
-
},
|
| 2440 |
-
{
|
| 2441 |
-
"cell_type": "markdown",
|
| 2442 |
-
"metadata": {},
|
| 2443 |
-
"source": [
|
| 2444 |
-
"## <a id='toc3_3_'></a>[Random Baseline](#toc0_)"
|
| 2445 |
]
|
| 2446 |
},
|
| 2447 |
{
|
|
|
|
| 5 |
"metadata": {},
|
| 6 |
"source": [
|
| 7 |
"**Table of contents**<a id='toc0_'></a> \n",
|
| 8 |
+
"- [Sources](#toc1_) \n",
|
| 9 |
"- [Load and import](#toc2_) \n",
|
| 10 |
" - [Import libraries](#toc2_1_) \n",
|
| 11 |
" - [Load configuration](#toc2_2_) \n",
|
| 12 |
+
" - [Load, split, export data](#toc2_3_) \n",
|
| 13 |
"- [EDA](#toc3_) \n",
|
| 14 |
+
" - [Classes distribution](#toc3_1_) \n",
|
| 15 |
+
" - [🚧 Constats :](#toc3_2_) \n",
|
| 16 |
+
" - [🚧 Image exploration](#toc3_3_) \n",
|
| 17 |
+
"- [Data preprocessing](#toc4_) \n",
|
| 18 |
+
"- [🦄🦄 CHECKPOINT 🦄🦄](#toc5_) \n",
|
| 19 |
+
"- [Model Training](#toc6_) \n",
|
| 20 |
+
"- [🚧 fin Code JL](#toc7_) \n",
|
| 21 |
+
" - [Random Baseline](#toc7_1_) \n",
|
| 22 |
"\n",
|
| 23 |
"<!-- vscode-jupyter-toc-config\n",
|
| 24 |
"\tnumbering=false\n",
|
|
|
|
| 34 |
"cell_type": "markdown",
|
| 35 |
"metadata": {},
|
| 36 |
"source": [
|
| 37 |
+
"# <a id='toc1_'></a>[Sources](#toc0_)\n",
|
| 38 |
"\n",
|
| 39 |
"https://huggingface.co/datasets/pyronear/pyro-sdis\n",
|
| 40 |
"\n",
|
|
|
|
| 66 |
"text": [
|
| 67 |
"/Users/julmat/Documents/hugging_face/frugal_cviz/.venv/lib/python3.12/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
|
| 68 |
" from .autonotebook import tqdm as notebook_tqdm\n",
|
| 69 |
+
"[codecarbon WARNING @ 10:42:32] Multiple instances of codecarbon are allowed to run at the same time.\n",
|
| 70 |
+
"[codecarbon INFO @ 10:42:32] [setup] RAM Tracking...\n",
|
| 71 |
+
"[codecarbon INFO @ 10:42:32] [setup] CPU Tracking...\n",
|
| 72 |
+
"[codecarbon WARNING @ 10:42:33] No CPU tracking mode found. Falling back on CPU constant mode. \n",
|
| 73 |
" Mac OS and ARM processor detected: Please enable PowerMetrics sudo to measure CPU\n",
|
| 74 |
"\n",
|
| 75 |
+
"[codecarbon INFO @ 10:42:33] CPU Model on constant consumption mode: Apple M1\n",
|
| 76 |
+
"[codecarbon INFO @ 10:42:33] [setup] GPU Tracking...\n",
|
| 77 |
+
"[codecarbon INFO @ 10:42:33] No GPU found.\n",
|
| 78 |
+
"[codecarbon INFO @ 10:42:33] >>> Tracker's metadata:\n",
|
| 79 |
+
"[codecarbon INFO @ 10:42:33] Platform system: macOS-15.2-arm64-arm-64bit\n",
|
| 80 |
+
"[codecarbon INFO @ 10:42:33] Python version: 3.12.7\n",
|
| 81 |
+
"[codecarbon INFO @ 10:42:33] CodeCarbon version: 2.8.3\n",
|
| 82 |
+
"[codecarbon INFO @ 10:42:33] Available RAM : 16.000 GB\n",
|
| 83 |
+
"[codecarbon INFO @ 10:42:33] CPU count: 8\n",
|
| 84 |
+
"[codecarbon INFO @ 10:42:33] CPU model: Apple M1\n",
|
| 85 |
+
"[codecarbon INFO @ 10:42:33] GPU count: None\n",
|
| 86 |
+
"[codecarbon INFO @ 10:42:33] GPU model: None\n",
|
| 87 |
+
"[codecarbon INFO @ 10:42:34] Saving emissions data to file /Users/julmat/Documents/hugging_face/frugal_cviz/emissions.csv\n"
|
| 88 |
]
|
| 89 |
}
|
| 90 |
],
|
|
|
|
| 93 |
"from fastapi import APIRouter\n",
|
| 94 |
"import logging\n",
|
| 95 |
"import matplotlib.pyplot as plt\n",
|
| 96 |
+
"import os\n",
|
| 97 |
"from PIL import Image, ImageOps, ImageEnhance, ImageFilter\n",
|
| 98 |
"import plotly.express as px\n",
|
| 99 |
"import random\n",
|
|
|
|
| 102 |
"# ML\n",
|
| 103 |
"from keras import Model\n",
|
| 104 |
"from keras.applications import EfficientNetB0\n",
|
|
|
|
| 105 |
"from keras.layers import Flatten, Dense\n",
|
| 106 |
+
"from keras.metrics import Precision, Recall\n",
|
| 107 |
"from keras.optimizers import AdamW\n",
|
| 108 |
+
"from keras.utils import image_dataset_from_directory\n",
|
| 109 |
"import numpy as np\n",
|
| 110 |
"import pandas as pd\n",
|
| 111 |
"from scipy.ndimage import median_filter\n",
|
|
|
|
| 120 |
"from tasks.utils.emissions import tracker, clean_emissions_data, get_space_info\n",
|
| 121 |
"from tasks.image import parse_boxes, compute_iou, compute_max_iou\n",
|
| 122 |
"\n",
|
| 123 |
+
"\n",
|
| 124 |
"# Logging configuration (see all outputs, even DEBUG or INFO)\n",
|
| 125 |
"logger = logging.getLogger()\n",
|
| 126 |
"logger.setLevel(logging.INFO)"
|
|
|
|
| 142 |
"# local config\n",
|
| 143 |
"with open(\"config.yaml\", \"r\") as f:\n",
|
| 144 |
" cfg = yaml.safe_load(f)\n",
|
| 145 |
+
"# Data\n",
|
| 146 |
"OUTPUT_DIR = cfg[\"data_root_dir\"]\n",
|
| 147 |
"DB_INFO_URI = cfg[\"db_info_uri\"]\n",
|
| 148 |
"REPO_ID = cfg[\"repo_id\"]\n",
|
| 149 |
"SPLIT_SIZE = cfg[\"split_size\"]\n",
|
| 150 |
"RDM_SEED = cfg[\"rdm_seed\"]\n",
|
| 151 |
+
"# Model (common)\n",
|
| 152 |
+
"MODELS_ROOT_DIR = cfg[\"models_common\"][\"models_root_dir\"]\n",
|
| 153 |
+
"CLASSES = cfg[\"models_common\"][\"classes\"]\n",
|
| 154 |
+
"LOG_DIR = cfg[\"models_common\"][\"log_dir\"]\n",
|
| 155 |
+
"CHKPT_DIR = cfg[\"models_common\"][\"chkpts_dir\"]\n",
|
| 156 |
"\n",
|
| 157 |
"request = ImageEvaluationRequest()"
|
| 158 |
]
|
|
|
|
| 785 |
"cell_type": "markdown",
|
| 786 |
"metadata": {},
|
| 787 |
"source": [
|
| 788 |
+
"## <a id='toc3_1_'></a>[Classes distribution](#toc0_)"
|
| 789 |
]
|
| 790 |
},
|
| 791 |
{
|
|
|
|
| 1756 |
"cell_type": "markdown",
|
| 1757 |
"metadata": {},
|
| 1758 |
"source": [
|
| 1759 |
+
"## <a id='toc3_2_'></a>[🚧 Constats :](#toc0_)\n",
|
| 1760 |
"\n",
|
| 1761 |
"- 1 seul définition : 1280x720\n",
|
| 1762 |
"- 1 seul format : JPG\n",
|
| 1763 |
"- 1 seul mode : RGB\n",
|
| 1764 |
"\n",
|
| 1765 |
+
"- voir répartition des caméras et partenaires selon les splits"
|
| 1766 |
]
|
| 1767 |
},
|
| 1768 |
{
|
| 1769 |
"cell_type": "markdown",
|
| 1770 |
"metadata": {},
|
| 1771 |
"source": [
|
| 1772 |
+
"## <a id='toc3_3_'></a>[🚧 Image exploration](#toc0_)"
|
| 1773 |
]
|
| 1774 |
},
|
| 1775 |
{
|
|
|
|
| 1920 |
"cell_type": "markdown",
|
| 1921 |
"metadata": {},
|
| 1922 |
"source": [
|
| 1923 |
+
"# <a id='toc4_'></a>[Data preprocessing](#toc0_)"
|
| 1924 |
]
|
| 1925 |
},
|
| 1926 |
{
|
|
|
|
| 1955 |
"name": "stderr",
|
| 1956 |
"output_type": "stream",
|
| 1957 |
"text": [
|
| 1958 |
+
"INFO:root:data/keras already exists: data already formatted\n"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1959 |
]
|
| 1960 |
}
|
| 1961 |
],
|
|
|
|
| 1967 |
"cell_type": "markdown",
|
| 1968 |
"metadata": {},
|
| 1969 |
"source": [
|
| 1970 |
+
"Balance classes by oversampling \"no_smoke\" label"
|
| 1971 |
+
]
|
| 1972 |
+
},
|
| 1973 |
+
{
|
| 1974 |
+
"cell_type": "markdown",
|
| 1975 |
+
"metadata": {},
|
| 1976 |
+
"source": [
|
| 1977 |
+
"# <a id='toc5_'></a>[🦄🦄 CHECKPOINT 🦄🦄](#toc0_)"
|
| 1978 |
+
]
|
| 1979 |
+
},
|
| 1980 |
+
{
|
| 1981 |
+
"cell_type": "markdown",
|
| 1982 |
+
"metadata": {},
|
| 1983 |
+
"source": [
|
| 1984 |
+
"# 🚧 OVERSAMPLING"
|
| 1985 |
+
]
|
| 1986 |
+
},
|
| 1987 |
+
{
|
| 1988 |
+
"cell_type": "code",
|
| 1989 |
+
"execution_count": 72,
|
| 1990 |
+
"metadata": {},
|
| 1991 |
+
"outputs": [
|
| 1992 |
+
{
|
| 1993 |
+
"data": {
|
| 1994 |
+
"text/html": [
|
| 1995 |
+
"<div>\n",
|
| 1996 |
+
"<style scoped>\n",
|
| 1997 |
+
" .dataframe tbody tr th:only-of-type {\n",
|
| 1998 |
+
" vertical-align: middle;\n",
|
| 1999 |
+
" }\n",
|
| 2000 |
+
"\n",
|
| 2001 |
+
" .dataframe tbody tr th {\n",
|
| 2002 |
+
" vertical-align: top;\n",
|
| 2003 |
+
" }\n",
|
| 2004 |
+
"\n",
|
| 2005 |
+
" .dataframe thead th {\n",
|
| 2006 |
+
" text-align: right;\n",
|
| 2007 |
+
" }\n",
|
| 2008 |
+
"</style>\n",
|
| 2009 |
+
"<table border=\"1\" class=\"dataframe\">\n",
|
| 2010 |
+
" <thead>\n",
|
| 2011 |
+
" <tr style=\"text-align: right;\">\n",
|
| 2012 |
+
" <th></th>\n",
|
| 2013 |
+
" <th>split</th>\n",
|
| 2014 |
+
" <th>label</th>\n",
|
| 2015 |
+
" <th>count</th>\n",
|
| 2016 |
+
" </tr>\n",
|
| 2017 |
+
" </thead>\n",
|
| 2018 |
+
" <tbody>\n",
|
| 2019 |
+
" <tr>\n",
|
| 2020 |
+
" <th>0</th>\n",
|
| 2021 |
+
" <td>test</td>\n",
|
| 2022 |
+
" <td>no_smoke</td>\n",
|
| 2023 |
+
" <td>967</td>\n",
|
| 2024 |
+
" </tr>\n",
|
| 2025 |
+
" <tr>\n",
|
| 2026 |
+
" <th>1</th>\n",
|
| 2027 |
+
" <td>test</td>\n",
|
| 2028 |
+
" <td>smoke</td>\n",
|
| 2029 |
+
" <td>4941</td>\n",
|
| 2030 |
+
" </tr>\n",
|
| 2031 |
+
" <tr>\n",
|
| 2032 |
+
" <th>2</th>\n",
|
| 2033 |
+
" <td>train</td>\n",
|
| 2034 |
+
" <td>no_smoke</td>\n",
|
| 2035 |
+
" <td>3802</td>\n",
|
| 2036 |
+
" </tr>\n",
|
| 2037 |
+
" <tr>\n",
|
| 2038 |
+
" <th>3</th>\n",
|
| 2039 |
+
" <td>train</td>\n",
|
| 2040 |
+
" <td>smoke</td>\n",
|
| 2041 |
+
" <td>19827</td>\n",
|
| 2042 |
+
" </tr>\n",
|
| 2043 |
+
" </tbody>\n",
|
| 2044 |
+
"</table>\n",
|
| 2045 |
+
"</div>"
|
| 2046 |
+
],
|
| 2047 |
+
"text/plain": [
|
| 2048 |
+
" split label count\n",
|
| 2049 |
+
"0 test no_smoke 967\n",
|
| 2050 |
+
"1 test smoke 4941\n",
|
| 2051 |
+
"2 train no_smoke 3802\n",
|
| 2052 |
+
"3 train smoke 19827"
|
| 2053 |
+
]
|
| 2054 |
+
},
|
| 2055 |
+
"metadata": {},
|
| 2056 |
+
"output_type": "display_data"
|
| 2057 |
+
},
|
| 2058 |
+
{
|
| 2059 |
+
"data": {
|
| 2060 |
+
"text/html": [
|
| 2061 |
+
"<div>\n",
|
| 2062 |
+
"<style scoped>\n",
|
| 2063 |
+
" .dataframe tbody tr th:only-of-type {\n",
|
| 2064 |
+
" vertical-align: middle;\n",
|
| 2065 |
+
" }\n",
|
| 2066 |
+
"\n",
|
| 2067 |
+
" .dataframe tbody tr th {\n",
|
| 2068 |
+
" vertical-align: top;\n",
|
| 2069 |
+
" }\n",
|
| 2070 |
+
"\n",
|
| 2071 |
+
" .dataframe thead th {\n",
|
| 2072 |
+
" text-align: right;\n",
|
| 2073 |
+
" }\n",
|
| 2074 |
+
"</style>\n",
|
| 2075 |
+
"<table border=\"1\" class=\"dataframe\">\n",
|
| 2076 |
+
" <thead>\n",
|
| 2077 |
+
" <tr style=\"text-align: right;\">\n",
|
| 2078 |
+
" <th></th>\n",
|
| 2079 |
+
" <th>name</th>\n",
|
| 2080 |
+
" <th>label</th>\n",
|
| 2081 |
+
" <th>split</th>\n",
|
| 2082 |
+
" <th>uri</th>\n",
|
| 2083 |
+
" </tr>\n",
|
| 2084 |
+
" </thead>\n",
|
| 2085 |
+
" <tbody>\n",
|
| 2086 |
+
" <tr>\n",
|
| 2087 |
+
" <th>0</th>\n",
|
| 2088 |
+
" <td>sdis-07_marguerite-282_2024-01-30T17-01-57</td>\n",
|
| 2089 |
+
" <td>no_smoke</td>\n",
|
| 2090 |
+
" <td>train</td>\n",
|
| 2091 |
+
" <td>data/raw/images/train/sdis-07_marguerite-282_2...</td>\n",
|
| 2092 |
+
" </tr>\n",
|
| 2093 |
+
" <tr>\n",
|
| 2094 |
+
" <th>1</th>\n",
|
| 2095 |
+
" <td>force-06_courmettes-212_2024-01-04T12-06-07</td>\n",
|
| 2096 |
+
" <td>smoke</td>\n",
|
| 2097 |
+
" <td>train</td>\n",
|
| 2098 |
+
" <td>data/raw/images/train/force-06_courmettes-212_...</td>\n",
|
| 2099 |
+
" </tr>\n",
|
| 2100 |
+
" <tr>\n",
|
| 2101 |
+
" <th>2</th>\n",
|
| 2102 |
+
" <td>force-06_courmettes-212_2024-01-16T09-53-11</td>\n",
|
| 2103 |
+
" <td>no_smoke</td>\n",
|
| 2104 |
+
" <td>train</td>\n",
|
| 2105 |
+
" <td>data/raw/images/train/force-06_courmettes-212_...</td>\n",
|
| 2106 |
+
" </tr>\n",
|
| 2107 |
+
" <tr>\n",
|
| 2108 |
+
" <th>3</th>\n",
|
| 2109 |
+
" <td>force-06_courmettes-160_2024-04-26T09-19-42</td>\n",
|
| 2110 |
+
" <td>no_smoke</td>\n",
|
| 2111 |
+
" <td>train</td>\n",
|
| 2112 |
+
" <td>data/raw/images/train/force-06_courmettes-160_...</td>\n",
|
| 2113 |
+
" </tr>\n",
|
| 2114 |
+
" <tr>\n",
|
| 2115 |
+
" <th>4</th>\n",
|
| 2116 |
+
" <td>force-06_courmettes-212_2024-01-12T12-47-33</td>\n",
|
| 2117 |
+
" <td>no_smoke</td>\n",
|
| 2118 |
+
" <td>train</td>\n",
|
| 2119 |
+
" <td>data/raw/images/train/force-06_courmettes-212_...</td>\n",
|
| 2120 |
+
" </tr>\n",
|
| 2121 |
+
" <tr>\n",
|
| 2122 |
+
" <th>...</th>\n",
|
| 2123 |
+
" <td>...</td>\n",
|
| 2124 |
+
" <td>...</td>\n",
|
| 2125 |
+
" <td>...</td>\n",
|
| 2126 |
+
" <td>...</td>\n",
|
| 2127 |
+
" </tr>\n",
|
| 2128 |
+
" <tr>\n",
|
| 2129 |
+
" <th>33631</th>\n",
|
| 2130 |
+
" <td>sdis-07_brison-200_2024-01-04T12-49-23</td>\n",
|
| 2131 |
+
" <td>smoke</td>\n",
|
| 2132 |
+
" <td>test</td>\n",
|
| 2133 |
+
" <td>data/raw/images/test/sdis-07_brison-200_2024-0...</td>\n",
|
| 2134 |
+
" </tr>\n",
|
| 2135 |
+
" <tr>\n",
|
| 2136 |
+
" <th>33632</th>\n",
|
| 2137 |
+
" <td>sdis-07_brison-110_2024-03-08T11-23-33</td>\n",
|
| 2138 |
+
" <td>smoke</td>\n",
|
| 2139 |
+
" <td>test</td>\n",
|
| 2140 |
+
" <td>data/raw/images/test/sdis-07_brison-110_2024-0...</td>\n",
|
| 2141 |
+
" </tr>\n",
|
| 2142 |
+
" <tr>\n",
|
| 2143 |
+
" <th>33633</th>\n",
|
| 2144 |
+
" <td>sdis-07_marguerite-29_2024-01-31T12-18-29</td>\n",
|
| 2145 |
+
" <td>smoke</td>\n",
|
| 2146 |
+
" <td>test</td>\n",
|
| 2147 |
+
" <td>data/raw/images/test/sdis-07_marguerite-29_202...</td>\n",
|
| 2148 |
+
" </tr>\n",
|
| 2149 |
+
" <tr>\n",
|
| 2150 |
+
" <th>33634</th>\n",
|
| 2151 |
+
" <td>sdis-07_marguerite-29_2024-01-31T09-19-45</td>\n",
|
| 2152 |
+
" <td>smoke</td>\n",
|
| 2153 |
+
" <td>test</td>\n",
|
| 2154 |
+
" <td>data/raw/images/test/sdis-07_marguerite-29_202...</td>\n",
|
| 2155 |
+
" </tr>\n",
|
| 2156 |
+
" <tr>\n",
|
| 2157 |
+
" <th>33635</th>\n",
|
| 2158 |
+
" <td>sdis-07_marguerite-29_2024-03-25T10-11-12</td>\n",
|
| 2159 |
+
" <td>smoke</td>\n",
|
| 2160 |
+
" <td>test</td>\n",
|
| 2161 |
+
" <td>data/raw/images/test/sdis-07_marguerite-29_202...</td>\n",
|
| 2162 |
+
" </tr>\n",
|
| 2163 |
+
" </tbody>\n",
|
| 2164 |
+
"</table>\n",
|
| 2165 |
+
"<p>33636 rows × 4 columns</p>\n",
|
| 2166 |
+
"</div>"
|
| 2167 |
+
],
|
| 2168 |
+
"text/plain": [
|
| 2169 |
+
" name label split \\\n",
|
| 2170 |
+
"0 sdis-07_marguerite-282_2024-01-30T17-01-57 no_smoke train \n",
|
| 2171 |
+
"1 force-06_courmettes-212_2024-01-04T12-06-07 smoke train \n",
|
| 2172 |
+
"2 force-06_courmettes-212_2024-01-16T09-53-11 no_smoke train \n",
|
| 2173 |
+
"3 force-06_courmettes-160_2024-04-26T09-19-42 no_smoke train \n",
|
| 2174 |
+
"4 force-06_courmettes-212_2024-01-12T12-47-33 no_smoke train \n",
|
| 2175 |
+
"... ... ... ... \n",
|
| 2176 |
+
"33631 sdis-07_brison-200_2024-01-04T12-49-23 smoke test \n",
|
| 2177 |
+
"33632 sdis-07_brison-110_2024-03-08T11-23-33 smoke test \n",
|
| 2178 |
+
"33633 sdis-07_marguerite-29_2024-01-31T12-18-29 smoke test \n",
|
| 2179 |
+
"33634 sdis-07_marguerite-29_2024-01-31T09-19-45 smoke test \n",
|
| 2180 |
+
"33635 sdis-07_marguerite-29_2024-03-25T10-11-12 smoke test \n",
|
| 2181 |
+
"\n",
|
| 2182 |
+
" uri \n",
|
| 2183 |
+
"0 data/raw/images/train/sdis-07_marguerite-282_2... \n",
|
| 2184 |
+
"1 data/raw/images/train/force-06_courmettes-212_... \n",
|
| 2185 |
+
"2 data/raw/images/train/force-06_courmettes-212_... \n",
|
| 2186 |
+
"3 data/raw/images/train/force-06_courmettes-160_... \n",
|
| 2187 |
+
"4 data/raw/images/train/force-06_courmettes-212_... \n",
|
| 2188 |
+
"... ... \n",
|
| 2189 |
+
"33631 data/raw/images/test/sdis-07_brison-200_2024-0... \n",
|
| 2190 |
+
"33632 data/raw/images/test/sdis-07_brison-110_2024-0... \n",
|
| 2191 |
+
"33633 data/raw/images/test/sdis-07_marguerite-29_202... \n",
|
| 2192 |
+
"33634 data/raw/images/test/sdis-07_marguerite-29_202... \n",
|
| 2193 |
+
"33635 data/raw/images/test/sdis-07_marguerite-29_202... \n",
|
| 2194 |
+
"\n",
|
| 2195 |
+
"[33636 rows x 4 columns]"
|
| 2196 |
+
]
|
| 2197 |
+
},
|
| 2198 |
+
"metadata": {},
|
| 2199 |
+
"output_type": "display_data"
|
| 2200 |
+
}
|
| 2201 |
+
],
|
| 2202 |
+
"source": [
|
| 2203 |
+
"count_df_ = df.groupby([\"split\", \"label\"]).size().reset_index(name=\"count\")\n",
|
| 2204 |
+
"count_df_ = count_df_.loc[count_df_[\"split\"] != \"val\"]\n",
|
| 2205 |
+
"display(count_df_)\n",
|
| 2206 |
+
"display(df)\n",
|
| 2207 |
+
"\n",
|
| 2208 |
+
"\n",
|
| 2209 |
+
"# for split in [\"train\", \"test\"]\n",
|
| 2210 |
+
" # quel label pour le min count ?\n",
|
| 2211 |
+
" # calculer l'écart min / max\n",
|
| 2212 |
+
" # échantillonner le DF pour ce split et ce label\n",
|
| 2213 |
+
" # boucler sur les lignes jusqu'à complétion de l'écart\n",
|
| 2214 |
+
" # avec une technique de data aug différente pour chaque époque => fonction"
|
| 2215 |
]
|
| 2216 |
},
|
| 2217 |
{
|
|
|
|
| 2239 |
"cell_type": "markdown",
|
| 2240 |
"metadata": {},
|
| 2241 |
"source": [
|
| 2242 |
+
"# <a id='toc6_'></a>[Model Training](#toc0_)"
|
| 2243 |
+
]
|
| 2244 |
+
},
|
| 2245 |
+
{
|
| 2246 |
+
"cell_type": "markdown",
|
| 2247 |
+
"metadata": {},
|
| 2248 |
+
"source": [
|
| 2249 |
+
"## Load configuration"
|
| 2250 |
+
]
|
| 2251 |
+
},
|
| 2252 |
+
{
|
| 2253 |
+
"cell_type": "markdown",
|
| 2254 |
+
"metadata": {},
|
| 2255 |
+
"source": [
|
| 2256 |
+
"# 🚧 MODEL CHOICE\n",
|
| 2257 |
+
"\n",
|
| 2258 |
+
"- https://paperswithcode.com/sota/image-classification-on-imagenet\n",
|
| 2259 |
+
"- https://keras.io/api/applications/efficientnet_v2/#efficientnetv2m-function"
|
| 2260 |
]
|
| 2261 |
},
|
| 2262 |
{
|
| 2263 |
"cell_type": "code",
|
| 2264 |
+
"execution_count": null,
|
| 2265 |
+
"metadata": {},
|
| 2266 |
+
"outputs": [],
|
| 2267 |
+
"source": [
|
| 2268 |
+
"# from keras.applications import EfficientNetV2M\n",
|
| 2269 |
+
"\n",
|
| 2270 |
+
"# model = EfficientNetV2M("
|
| 2271 |
+
]
|
| 2272 |
+
},
|
| 2273 |
+
{
|
| 2274 |
+
"cell_type": "code",
|
| 2275 |
+
"execution_count": 29,
|
| 2276 |
+
"metadata": {},
|
| 2277 |
+
"outputs": [],
|
| 2278 |
+
"source": [
|
| 2279 |
+
"# Model config\n",
|
| 2280 |
+
"model_name = \"EfficientNetB0\"\n",
|
| 2281 |
+
"input_size = (224, 224)\n",
|
| 2282 |
+
"batch_size = 48\n",
|
| 2283 |
+
"n_epochs = 100\n",
|
| 2284 |
+
"optimizer = AdamW(learning_rate=0.0002, weight_decay=0.05)\n",
|
| 2285 |
+
"loss = \"binary_crossentropy\"\n",
|
| 2286 |
+
"metrics = [\"accuracy\", Precision(), Recall()]\n",
|
| 2287 |
+
"# metrics = [\"accuracy\", Precision(), Recall(), \"f1_score\"]\n",
|
| 2288 |
+
"\n",
|
| 2289 |
+
"# Input paths\n",
|
| 2290 |
+
"data_dir = os.path.join(OUTPUT_DIR, \"keras\")\n",
|
| 2291 |
+
"train_dir = os.path.join(data_dir, \"train\")\n",
|
| 2292 |
+
"val_dir = os.path.join(data_dir, \"val\")\n",
|
| 2293 |
+
"test_dir = os.path.join(data_dir, \"test\")\n",
|
| 2294 |
+
"# Output paths\n",
|
| 2295 |
+
"MODEL_PATH = os.path.join(MODELS_ROOT_DIR, model_name)\n",
|
| 2296 |
+
"LOG_PATH = os.path.join(MODEL_PATH, LOG_DIR)\n",
|
| 2297 |
+
"CHKPT_PATH = os.path.join(MODEL_PATH, CHKPT_DIR)"
|
| 2298 |
+
]
|
| 2299 |
+
},
|
| 2300 |
+
{
|
| 2301 |
+
"cell_type": "markdown",
|
| 2302 |
+
"metadata": {},
|
| 2303 |
+
"source": [
|
| 2304 |
+
"## Prepare data"
|
| 2305 |
+
]
|
| 2306 |
+
},
|
| 2307 |
+
{
|
| 2308 |
+
"cell_type": "markdown",
|
| 2309 |
+
"metadata": {},
|
| 2310 |
+
"source": [
|
| 2311 |
+
"Create datasets from local images and labels"
|
| 2312 |
+
]
|
| 2313 |
+
},
|
| 2314 |
+
{
|
| 2315 |
+
"cell_type": "code",
|
| 2316 |
+
"execution_count": 30,
|
| 2317 |
"metadata": {},
|
| 2318 |
"outputs": [
|
| 2319 |
{
|
| 2320 |
"name": "stdout",
|
| 2321 |
"output_type": "stream",
|
| 2322 |
"text": [
|
| 2323 |
+
"Train dataset:\n",
|
| 2324 |
+
"Found 23629 files belonging to 2 classes.\n",
|
| 2325 |
+
"\n",
|
| 2326 |
+
"Val dataset:\n",
|
| 2327 |
+
"Found 4099 files belonging to 2 classes.\n",
|
| 2328 |
+
"\n",
|
| 2329 |
+
"Test dataset:\n",
|
| 2330 |
+
"Found 5908 files belonging to 2 classes.\n"
|
| 2331 |
]
|
| 2332 |
}
|
| 2333 |
],
|
| 2334 |
"source": [
|
| 2335 |
+
"# Prepare for outputs\n",
|
| 2336 |
+
"os.makedirs(MODELS_ROOT_DIR, exist_ok=True)\n",
|
| 2337 |
+
"y = df[\"label\"]\n",
|
| 2338 |
+
"X = df[\"uri\"]\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2339 |
"\n",
|
| 2340 |
+
"# Create datasets\n",
|
| 2341 |
+
"print(\"Train dataset:\")\n",
|
| 2342 |
+
"train_ds = image_dataset_from_directory(\n",
|
| 2343 |
+
" train_dir,\n",
|
| 2344 |
+
" labels=\"inferred\", # class names upon folders structure\n",
|
| 2345 |
+
" label_mode=\"int\", # integer encoding\n",
|
| 2346 |
+
" shuffle=True, # shuffle images\n",
|
| 2347 |
+
" seed=42, # random seed\n",
|
| 2348 |
+
" image_size=input_size, # automatic resizing\n",
|
| 2349 |
+
" batch_size=batch_size, # tensor shape[0]\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2350 |
")\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2351 |
"\n",
|
| 2352 |
+
"print(\"\\nVal dataset:\")\n",
|
| 2353 |
+
"val_ds = image_dataset_from_directory(\n",
|
| 2354 |
+
" val_dir,\n",
|
| 2355 |
" labels=\"inferred\", # class names upon folders structure\n",
|
| 2356 |
" label_mode=\"int\", # integer encoding\n",
|
| 2357 |
+
" shuffle=True, # shuffle images\n",
|
|
|
|
|
|
|
| 2358 |
" seed=42, # random seed\n",
|
| 2359 |
" image_size=input_size, # automatic resizing\n",
|
| 2360 |
" batch_size=batch_size, # tensor shape[0]\n",
|
|
|
|
| 2362 |
"\n",
|
| 2363 |
"print(\"\\nTest dataset:\")\n",
|
| 2364 |
"test_ds = image_dataset_from_directory(\n",
|
| 2365 |
+
" test_dir,\n",
|
| 2366 |
" labels=\"inferred\", # class names upon folders structure\n",
|
| 2367 |
" label_mode=\"int\", # integer encoding\n",
|
| 2368 |
+
" shuffle=False, # do not shuffle images\n",
|
| 2369 |
" seed=42, # random seed\n",
|
| 2370 |
" image_size=input_size, # automatic resizing\n",
|
| 2371 |
" batch_size=batch_size, # tensor shape[0]\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2372 |
")"
|
| 2373 |
]
|
| 2374 |
},
|
|
|
|
| 2376 |
"cell_type": "markdown",
|
| 2377 |
"metadata": {},
|
| 2378 |
"source": [
|
| 2379 |
+
"## Prepare model"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2380 |
]
|
| 2381 |
},
|
| 2382 |
{
|
| 2383 |
"cell_type": "code",
|
| 2384 |
+
"execution_count": 31,
|
| 2385 |
"metadata": {},
|
| 2386 |
"outputs": [
|
| 2387 |
{
|
| 2388 |
"name": "stderr",
|
| 2389 |
"output_type": "stream",
|
| 2390 |
"text": [
|
| 2391 |
+
"/Users/julmat/Documents/hugging_face/frugal_cviz/.venv/lib/python3.12/site-packages/keras/src/layers/layer.py:393: UserWarning:\n",
|
| 2392 |
+
"\n",
|
| 2393 |
+
"`build()` was called on layer 'conditional_augmentation_3', however the layer does not have a `build()` method implemented and it looks like it has unbuilt state. This will cause the layer to be marked as built, despite not being actually built, which may cause failures down the line. Make sure to implement a proper `build()` method.\n",
|
| 2394 |
+
"\n"
|
| 2395 |
]
|
| 2396 |
},
|
| 2397 |
{
|
| 2398 |
"data": {
|
| 2399 |
"text/html": [
|
| 2400 |
+
"<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"functional_3\"</span>\n",
|
| 2401 |
"</pre>\n"
|
| 2402 |
],
|
| 2403 |
"text/plain": [
|
| 2404 |
+
"\u001b[1mModel: \"functional_3\"\u001b[0m\n"
|
| 2405 |
]
|
| 2406 |
},
|
| 2407 |
"metadata": {},
|
|
|
|
| 2413 |
"<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
|
| 2414 |
"┃<span style=\"font-weight: bold\"> Layer (type) </span>┃<span style=\"font-weight: bold\"> Output Shape </span>┃<span style=\"font-weight: bold\"> Param # </span>┃\n",
|
| 2415 |
"┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
|
| 2416 |
+
"│ input_layer_7 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>) │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">224</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">224</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">3</span>) │ <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
|
| 2417 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2418 |
+
"│ conditional_augmentation_3 │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">224</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">224</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">3</span>) │ <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
|
| 2419 |
"│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">ConditionalAugmentation</span>) │ │ │\n",
|
| 2420 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2421 |
"│ efficientnetb0 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Functional</span>) │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">7</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">7</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">1280</span>) │ <span style=\"color: #00af00; text-decoration-color: #00af00\">4,049,571</span> │\n",
|
| 2422 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2423 |
+
"│ flatten_3 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Flatten</span>) │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">62720</span>) │ <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
|
| 2424 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2425 |
+
"│ dense_3 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>) │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">1</span>) │ <span style=\"color: #00af00; text-decoration-color: #00af00\">62,721</span> │\n",
|
| 2426 |
"└─────────────────────────────────┴────────────────────────┴───────────────┘\n",
|
| 2427 |
"</pre>\n"
|
| 2428 |
],
|
|
|
|
| 2430 |
"┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
|
| 2431 |
"┃\u001b[1m \u001b[0m\u001b[1mLayer (type) \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
|
| 2432 |
"┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
|
| 2433 |
+
"│ input_layer_7 (\u001b[38;5;33mInputLayer\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m224\u001b[0m, \u001b[38;5;34m224\u001b[0m, \u001b[38;5;34m3\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │\n",
|
| 2434 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2435 |
+
"│ conditional_augmentation_3 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m224\u001b[0m, \u001b[38;5;34m224\u001b[0m, \u001b[38;5;34m3\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │\n",
|
| 2436 |
"│ (\u001b[38;5;33mConditionalAugmentation\u001b[0m) │ │ │\n",
|
| 2437 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2438 |
"│ efficientnetb0 (\u001b[38;5;33mFunctional\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m7\u001b[0m, \u001b[38;5;34m7\u001b[0m, \u001b[38;5;34m1280\u001b[0m) │ \u001b[38;5;34m4,049,571\u001b[0m │\n",
|
| 2439 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2440 |
+
"│ flatten_3 (\u001b[38;5;33mFlatten\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m62720\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │\n",
|
| 2441 |
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
|
| 2442 |
+
"│ dense_3 (\u001b[38;5;33mDense\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m1\u001b[0m) │ \u001b[38;5;34m62,721\u001b[0m │\n",
|
| 2443 |
"└─────────────────────────────────┴────────────────────────┴───────────────┘\n"
|
| 2444 |
]
|
| 2445 |
},
|
|
|
|
| 2449 |
{
|
| 2450 |
"data": {
|
| 2451 |
"text/html": [
|
| 2452 |
+
"<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">4,112,292</span> (15.69 MB)\n",
|
| 2453 |
"</pre>\n"
|
| 2454 |
],
|
| 2455 |
"text/plain": [
|
| 2456 |
+
"\u001b[1m Total params: \u001b[0m\u001b[38;5;34m4,112,292\u001b[0m (15.69 MB)\n"
|
| 2457 |
]
|
| 2458 |
},
|
| 2459 |
"metadata": {},
|
|
|
|
| 2462 |
{
|
| 2463 |
"data": {
|
| 2464 |
"text/html": [
|
| 2465 |
+
"<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">62,721</span> (245.00 KB)\n",
|
| 2466 |
"</pre>\n"
|
| 2467 |
],
|
| 2468 |
"text/plain": [
|
| 2469 |
+
"\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m62,721\u001b[0m (245.00 KB)\n"
|
| 2470 |
]
|
| 2471 |
},
|
| 2472 |
"metadata": {},
|
|
|
|
| 2487 |
}
|
| 2488 |
],
|
| 2489 |
"source": [
|
| 2490 |
+
"# Load pre-trained model without top layers\n",
|
|
|
|
|
|
|
| 2491 |
"model = EfficientNetB0(\n",
|
| 2492 |
" weights=\"imagenet\", # pre-trained weights\n",
|
| 2493 |
" include_top=False, # no dense layer\n",
|
| 2494 |
" input_shape=(input_size[0], input_size[1], 3), # input shape\n",
|
| 2495 |
")\n",
|
| 2496 |
+
"# Create explicit input layer\n",
|
| 2497 |
"inputs = tf.keras.Input(shape=(input_size[0], input_size[1], 3))\n",
|
| 2498 |
"# add data augmentation\n",
|
| 2499 |
"augmented = ConditionalAugmentation(rate=0.4)(inputs)\n",
|
| 2500 |
"x = model(augmented)\n",
|
| 2501 |
"\n",
|
| 2502 |
+
"# Flatten output\n",
|
| 2503 |
"x = Flatten()(x)\n",
|
| 2504 |
+
"# For feature extraction only\n",
|
| 2505 |
"for layer in model.layers:\n",
|
| 2506 |
" layer.trainable = False\n",
|
| 2507 |
+
"# New FC layer for binary classification\n",
|
| 2508 |
+
"predictions = Dense(1, activation=\"sigmoid\")(x)\n",
|
| 2509 |
+
"# Define new model\n",
|
| 2510 |
"effnetB0 = Model(inputs=inputs, outputs=predictions)\n",
|
| 2511 |
+
"# Display model summary\n",
|
| 2512 |
"effnetB0.summary()"
|
| 2513 |
]
|
| 2514 |
},
|
|
|
|
| 2516 |
"cell_type": "markdown",
|
| 2517 |
"metadata": {},
|
| 2518 |
"source": [
|
| 2519 |
+
"## Training"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2520 |
]
|
| 2521 |
},
|
| 2522 |
{
|
| 2523 |
+
"cell_type": "markdown",
|
|
|
|
| 2524 |
"metadata": {},
|
|
|
|
| 2525 |
"source": [
|
| 2526 |
+
"Follow with : `tensorboard --logdir models/EfficientNetB0/runs`"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2527 |
]
|
| 2528 |
},
|
| 2529 |
{
|
| 2530 |
"cell_type": "code",
|
| 2531 |
+
"execution_count": 32,
|
| 2532 |
"metadata": {},
|
| 2533 |
+
"outputs": [
|
| 2534 |
+
{
|
| 2535 |
+
"name": "stderr",
|
| 2536 |
+
"output_type": "stream",
|
| 2537 |
+
"text": [
|
| 2538 |
+
"INFO:root:⚙️ compiling\n",
|
| 2539 |
+
"INFO:root:🛎️ declaring callbacks\n",
|
| 2540 |
+
"INFO:root:💪 starting training\n"
|
| 2541 |
+
]
|
| 2542 |
+
},
|
| 2543 |
+
{
|
| 2544 |
+
"name": "stdout",
|
| 2545 |
+
"output_type": "stream",
|
| 2546 |
+
"text": [
|
| 2547 |
+
"Epoch 1/100\n",
|
| 2548 |
+
"\u001b[1m493/493\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m406s\u001b[0m 815ms/step - accuracy: 0.8178 - loss: 0.4710 - precision_3: 0.8521 - recall_3: 0.9467 - val_accuracy: 0.7821 - val_loss: 0.5143 - val_precision_3: 0.8256 - val_recall_3: 0.9283\n",
|
| 2549 |
+
"Epoch 2/100\n",
|
| 2550 |
+
"\u001b[1m493/493\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m398s\u001b[0m 808ms/step - accuracy: 0.8334 - loss: 0.4258 - precision_3: 0.8697 - recall_3: 0.9422 - val_accuracy: 0.8046 - val_loss: 0.5463 - val_precision_3: 0.8188 - val_recall_3: 0.9757\n",
|
| 2551 |
+
"Epoch 3/100\n",
|
| 2552 |
+
"\u001b[1m493/493\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m394s\u001b[0m 799ms/step - accuracy: 0.8461 - loss: 0.4060 - precision_3: 0.8763 - recall_3: 0.9506 - val_accuracy: 0.7590 - val_loss: 0.6189 - val_precision_3: 0.8438 - val_recall_3: 0.8636\n",
|
| 2553 |
+
"Epoch 4/100\n",
|
| 2554 |
+
"\u001b[1m493/493\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m383s\u001b[0m 776ms/step - accuracy: 0.8472 - loss: 0.3938 - precision_3: 0.8793 - recall_3: 0.9476 - val_accuracy: 0.7785 - val_loss: 0.5448 - val_precision_3: 0.8263 - val_recall_3: 0.9214\n",
|
| 2555 |
+
"Epoch 5/100\n",
|
| 2556 |
+
"\u001b[1m493/493\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m361s\u001b[0m 733ms/step - accuracy: 0.8579 - loss: 0.3752 - precision_3: 0.8865 - recall_3: 0.9524 - val_accuracy: 0.7856 - val_loss: 0.6732 - val_precision_3: 0.8133 - val_recall_3: 0.9559\n",
|
| 2557 |
+
"Epoch 6/100\n",
|
| 2558 |
+
"\u001b[1m493/493\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m368s\u001b[0m 747ms/step - accuracy: 0.8504 - loss: 0.3840 - precision_3: 0.8866 - recall_3: 0.9417 - val_accuracy: 0.8104 - val_loss: 0.5844 - val_precision_3: 0.8233 - val_recall_3: 0.9766\n",
|
| 2559 |
+
"Epoch 7/100\n",
|
| 2560 |
+
"\u001b[1m493/493\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m366s\u001b[0m 743ms/step - accuracy: 0.8543 - loss: 0.3848 - precision_3: 0.8870 - recall_3: 0.9468 - val_accuracy: 0.7868 - val_loss: 0.6400 - val_precision_3: 0.8209 - val_recall_3: 0.9439\n",
|
| 2561 |
+
"Epoch 8/100\n",
|
| 2562 |
+
"\u001b[1m493/493\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m359s\u001b[0m 728ms/step - accuracy: 0.8572 - loss: 0.3700 - precision_3: 0.8884 - recall_3: 0.9486 - val_accuracy: 0.7965 - val_loss: 0.5693 - val_precision_3: 0.8252 - val_recall_3: 0.9514\n",
|
| 2563 |
+
"Epoch 9/100\n",
|
| 2564 |
+
"\u001b[1m493/493\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m363s\u001b[0m 737ms/step - accuracy: 0.8564 - loss: 0.3684 - precision_3: 0.8890 - recall_3: 0.9467 - val_accuracy: 0.7939 - val_loss: 0.6147 - val_precision_3: 0.8341 - val_recall_3: 0.9319\n",
|
| 2565 |
+
"Epoch 10/100\n",
|
| 2566 |
+
"\u001b[1m493/493\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m369s\u001b[0m 749ms/step - accuracy: 0.8603 - loss: 0.3629 - precision_3: 0.8904 - recall_3: 0.9500 - val_accuracy: 0.8026 - val_loss: 0.5798 - val_precision_3: 0.8311 - val_recall_3: 0.9505\n",
|
| 2567 |
+
"Epoch 11/100\n",
|
| 2568 |
+
"\u001b[1m493/493\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m368s\u001b[0m 746ms/step - accuracy: 0.8634 - loss: 0.3589 - precision_3: 0.8931 - recall_3: 0.9507 - val_accuracy: 0.7573 - val_loss: 0.6606 - val_precision_3: 0.8164 - val_recall_3: 0.9052\n"
|
| 2569 |
+
]
|
| 2570 |
+
},
|
| 2571 |
+
{
|
| 2572 |
+
"name": "stderr",
|
| 2573 |
+
"output_type": "stream",
|
| 2574 |
+
"text": [
|
| 2575 |
+
"INFO:root:🧐 evaluating model\n"
|
| 2576 |
+
]
|
| 2577 |
+
},
|
| 2578 |
+
{
|
| 2579 |
+
"name": "stdout",
|
| 2580 |
+
"output_type": "stream",
|
| 2581 |
+
"text": [
|
| 2582 |
+
"\u001b[1m124/124\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m72s\u001b[0m 579ms/step - accuracy: 0.6824 - loss: 0.8199 - precision_3: 0.6028 - recall_3: 0.7630\n",
|
| 2583 |
+
"\u001b[1m124/124\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m74s\u001b[0m 586ms/step\n"
|
| 2584 |
+
]
|
| 2585 |
+
},
|
| 2586 |
+
{
|
| 2587 |
+
"name": "stderr",
|
| 2588 |
+
"output_type": "stream",
|
| 2589 |
+
"text": [
|
| 2590 |
+
"INFO:root:📈 plotting results\n"
|
| 2591 |
+
]
|
| 2592 |
+
},
|
| 2593 |
+
{
|
| 2594 |
+
"name": "stdout",
|
| 2595 |
+
"output_type": "stream",
|
| 2596 |
+
"text": [
|
| 2597 |
+
" precision recall f1-score support\n",
|
| 2598 |
+
"\n",
|
| 2599 |
+
" no_smoke 0.16 1.00 0.28 967\n",
|
| 2600 |
+
" smoke 0.00 0.00 0.00 4941\n",
|
| 2601 |
+
"\n",
|
| 2602 |
+
" accuracy 0.16 5908\n",
|
| 2603 |
+
" macro avg 0.08 0.50 0.14 5908\n",
|
| 2604 |
+
"weighted avg 0.03 0.16 0.05 5908\n",
|
| 2605 |
+
"\n"
|
| 2606 |
+
]
|
| 2607 |
+
},
|
| 2608 |
+
{
|
| 2609 |
+
"name": "stderr",
|
| 2610 |
+
"output_type": "stream",
|
| 2611 |
+
"text": [
|
| 2612 |
+
"2025-01-27 12:32:42.266301: W tensorflow/core/framework/local_rendezvous.cc:404] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence\n",
|
| 2613 |
+
"/Users/julmat/Documents/hugging_face/frugal_cviz/.venv/lib/python3.12/site-packages/sklearn/metrics/_classification.py:1565: UndefinedMetricWarning:\n",
|
| 2614 |
+
"\n",
|
| 2615 |
+
"Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.\n",
|
| 2616 |
+
"\n",
|
| 2617 |
+
"/Users/julmat/Documents/hugging_face/frugal_cviz/.venv/lib/python3.12/site-packages/sklearn/metrics/_classification.py:1565: UndefinedMetricWarning:\n",
|
| 2618 |
+
"\n",
|
| 2619 |
+
"Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.\n",
|
| 2620 |
+
"\n",
|
| 2621 |
+
"/Users/julmat/Documents/hugging_face/frugal_cviz/.venv/lib/python3.12/site-packages/sklearn/metrics/_classification.py:1565: UndefinedMetricWarning:\n",
|
| 2622 |
+
"\n",
|
| 2623 |
+
"Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.\n",
|
| 2624 |
+
"\n",
|
| 2625 |
+
"/Users/julmat/Documents/hugging_face/frugal_cviz/.venv/lib/python3.12/site-packages/sklearn/metrics/_classification.py:1565: UndefinedMetricWarning:\n",
|
| 2626 |
+
"\n",
|
| 2627 |
+
"Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.\n",
|
| 2628 |
+
"\n",
|
| 2629 |
+
"/Users/julmat/Documents/hugging_face/frugal_cviz/.venv/lib/python3.12/site-packages/sklearn/metrics/_classification.py:1565: UndefinedMetricWarning:\n",
|
| 2630 |
+
"\n",
|
| 2631 |
+
"Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.\n",
|
| 2632 |
+
"\n",
|
| 2633 |
+
"/Users/julmat/Documents/hugging_face/frugal_cviz/.venv/lib/python3.12/site-packages/sklearn/metrics/_classification.py:1565: UndefinedMetricWarning:\n",
|
| 2634 |
+
"\n",
|
| 2635 |
+
"Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.\n",
|
| 2636 |
+
"\n"
|
| 2637 |
+
]
|
| 2638 |
+
},
|
| 2639 |
+
{
|
| 2640 |
+
"data": {
|
| 2641 |
+
"image/png": "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",
|
| 2642 |
+
"text/plain": [
|
| 2643 |
+
"<Figure size 600x400 with 2 Axes>"
|
| 2644 |
+
]
|
| 2645 |
+
},
|
| 2646 |
+
"metadata": {},
|
| 2647 |
+
"output_type": "display_data"
|
| 2648 |
+
},
|
| 2649 |
+
{
|
| 2650 |
+
"name": "stderr",
|
| 2651 |
+
"output_type": "stream",
|
| 2652 |
+
"text": [
|
| 2653 |
+
"INFO:root:📓 logging results\n"
|
| 2654 |
+
]
|
| 2655 |
+
}
|
| 2656 |
+
],
|
| 2657 |
"source": [
|
| 2658 |
+
"model_trained, history = eval_pretrained_model(\n",
|
| 2659 |
+
" model=effnetB0,\n",
|
| 2660 |
+
" train_ds=train_ds,\n",
|
| 2661 |
+
" val_ds=val_ds,\n",
|
| 2662 |
+
" test_ds=test_ds,\n",
|
| 2663 |
+
" LOG_DIR=LOG_PATH,\n",
|
| 2664 |
+
" CHKPT_DIR=CHKPT_PATH,\n",
|
| 2665 |
+
" model_name=model_name,\n",
|
| 2666 |
+
" input_size=input_size,\n",
|
| 2667 |
+
" batch_size=batch_size,\n",
|
| 2668 |
+
" n_epochs=n_epochs,\n",
|
| 2669 |
+
" optimizer=optimizer,\n",
|
| 2670 |
+
" loss=loss,\n",
|
| 2671 |
+
" metrics=metrics,\n",
|
| 2672 |
+
")"
|
| 2673 |
]
|
| 2674 |
},
|
| 2675 |
{
|
| 2676 |
"cell_type": "code",
|
| 2677 |
+
"execution_count": 33,
|
| 2678 |
"metadata": {},
|
| 2679 |
"outputs": [
|
| 2680 |
{
|
|
|
|
| 2684 |
"traceback": [
|
| 2685 |
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
|
| 2686 |
"\u001b[0;31mNameError\u001b[0m Traceback (most recent call last)",
|
| 2687 |
+
"Cell \u001b[0;32mIn[33], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m \u001b[43mstop\u001b[49m\n",
|
| 2688 |
"\u001b[0;31mNameError\u001b[0m: name 'stop' is not defined"
|
| 2689 |
]
|
| 2690 |
}
|
|
|
|
| 2737 |
"cell_type": "markdown",
|
| 2738 |
"metadata": {},
|
| 2739 |
"source": [
|
| 2740 |
+
"## <a id='toc7_1_'></a>[Random Baseline](#toc0_)"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2741 |
]
|
| 2742 |
},
|
| 2743 |
{
|
README.md
CHANGED
|
@@ -16,11 +16,12 @@ datasets:
|
|
| 16 |
|
| 17 |
# 🚧 TODO
|
| 18 |
|
|
|
|
|
|
|
| 19 |
- voir répartition partenaires, caméras, temporalité, annotations
|
| 20 |
-
- utiliser **classification binaire**
|
| 21 |
- métriques : **matrice de confusion** complète
|
| 22 |
- décrire voir **erreurs de types et conséquences**
|
| 23 |
-
- tester plusieurs pré-entraînements
|
| 24 |
- modèles :
|
| 25 |
- [ ] EfficientNet
|
| 26 |
- [ ] EfficientDet
|
|
|
|
| 16 |
|
| 17 |
# 🚧 TODO
|
| 18 |
|
| 19 |
+
- faire data augmentation en amont pour **no_smoke oversampling**
|
| 20 |
+
- réduire **n_epochs**
|
| 21 |
- voir répartition partenaires, caméras, temporalité, annotations
|
|
|
|
| 22 |
- métriques : **matrice de confusion** complète
|
| 23 |
- décrire voir **erreurs de types et conséquences**
|
| 24 |
+
- tester plusieurs pré-entraînements (est-ce que charger un modèle pré-entraîné ImageNet aide vraiment ?)
|
| 25 |
- modèles :
|
| 26 |
- [ ] EfficientNet
|
| 27 |
- [ ] EfficientDet
|
config.yaml
CHANGED
|
@@ -7,13 +7,20 @@ rdm_seed: 42
|
|
| 7 |
data_root_dir: "data"
|
| 8 |
raw_data_dir: "raw"
|
| 9 |
clr_hf_cache_script_abs_path: './src/clear_hf_cache.sh'
|
| 10 |
-
data_format: "keras" # "ultralytics" or "keras"
|
| 11 |
db_info_uri: "data_info.csv"
|
| 12 |
|
| 13 |
-
#
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
data_root_dir: "data"
|
| 8 |
raw_data_dir: "raw"
|
| 9 |
clr_hf_cache_script_abs_path: './src/clear_hf_cache.sh'
|
|
|
|
| 10 |
db_info_uri: "data_info.csv"
|
| 11 |
|
| 12 |
+
# Models
|
| 13 |
+
models_common:
|
| 14 |
+
models_root_dir: "models"
|
| 15 |
+
classes: ["no_smoke", "smoke"]
|
| 16 |
+
log_dir: "runs"
|
| 17 |
+
chkpts_dir : "chkpts"
|
| 18 |
+
# model_EffNet:
|
| 19 |
+
# model_name: "EfficientNetB0"
|
| 20 |
+
# input_size: (224, 224)
|
| 21 |
+
# batch_size: 48
|
| 22 |
+
# n_epochs: 100
|
| 23 |
+
# loss: "binary_crossentropy"
|
| 24 |
+
# metrics: ["accuracy"]
|
| 25 |
+
# optimizer: AdamW(learning_rate=0.0002, weight_decay=0.05)
|
| 26 |
+
# trained_model_uri: "EfficientNetB0_app.keras"
|
src/load_data.py
CHANGED
|
@@ -22,7 +22,6 @@ RDM_SEED = cfg["rdm_seed"]
|
|
| 22 |
OUTPUT_DIR = cfg["data_root_dir"]
|
| 23 |
RAW_DATA_DIR = os.path.join(OUTPUT_DIR, cfg["raw_data_dir"])
|
| 24 |
CLR_CACHE_SCRIPT = cfg["clr_hf_cache_script_abs_path"]
|
| 25 |
-
DATA_FORMAT = cfg["data_format"]
|
| 26 |
DB_INFO_URI = os.path.join(OUTPUT_DIR, cfg["db_info_uri"])
|
| 27 |
|
| 28 |
|
|
@@ -172,9 +171,13 @@ def format_data_keras(df):
|
|
| 172 |
if not os.path.exists(OUTPUT_DIR):
|
| 173 |
logging.warning(f"{OUTPUT_DIR} doesn't exist: (re)load data first")
|
| 174 |
return df
|
| 175 |
-
|
| 176 |
# Create Keras parent folder
|
| 177 |
keras_dir = os.path.join(OUTPUT_DIR, "keras")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 178 |
os.makedirs(keras_dir, exist_ok=True)
|
| 179 |
# Create splits folders
|
| 180 |
for split in df.split.unique():
|
|
@@ -195,6 +198,17 @@ def format_data_keras(df):
|
|
| 195 |
df.drop(columns="uri", inplace=True)
|
| 196 |
df.rename(columns={"uri_dest": "uri"}, inplace=True)
|
| 197 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 198 |
return df
|
| 199 |
|
| 200 |
|
|
|
|
| 22 |
OUTPUT_DIR = cfg["data_root_dir"]
|
| 23 |
RAW_DATA_DIR = os.path.join(OUTPUT_DIR, cfg["raw_data_dir"])
|
| 24 |
CLR_CACHE_SCRIPT = cfg["clr_hf_cache_script_abs_path"]
|
|
|
|
| 25 |
DB_INFO_URI = os.path.join(OUTPUT_DIR, cfg["db_info_uri"])
|
| 26 |
|
| 27 |
|
|
|
|
| 171 |
if not os.path.exists(OUTPUT_DIR):
|
| 172 |
logging.warning(f"{OUTPUT_DIR} doesn't exist: (re)load data first")
|
| 173 |
return df
|
| 174 |
+
|
| 175 |
# Create Keras parent folder
|
| 176 |
keras_dir = os.path.join(OUTPUT_DIR, "keras")
|
| 177 |
+
# Check if data already exists
|
| 178 |
+
if os.path.exists(keras_dir) and len(os.listdir("./data/keras")) > 0:
|
| 179 |
+
logging.info(f"{keras_dir} already exists: data already formatted")
|
| 180 |
+
return df
|
| 181 |
os.makedirs(keras_dir, exist_ok=True)
|
| 182 |
# Create splits folders
|
| 183 |
for split in df.split.unique():
|
|
|
|
| 198 |
df.drop(columns="uri", inplace=True)
|
| 199 |
df.rename(columns={"uri_dest": "uri"}, inplace=True)
|
| 200 |
|
| 201 |
+
return df
|
| 202 |
+
|
| 203 |
+
|
| 204 |
+
def oversample_class(df):
|
| 205 |
+
"""Oversample an under-represented class"""
|
| 206 |
+
count_df = df.groupby(["split", "label"]).size().reset_index(name="count")
|
| 207 |
+
count_df = count_df.loc[count_df["split"] != "val"]
|
| 208 |
+
|
| 209 |
+
|
| 210 |
+
|
| 211 |
+
|
| 212 |
return df
|
| 213 |
|
| 214 |
|