File size: 4,647 Bytes
3f75218
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
695636d
3f75218
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
af720a1
3f75218
 
af720a1
3f75218
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
af720a1
3f75218
 
 
 
 
 
 
a3e92cc
3f75218
 
d40ec4d
3f75218
a3e92cc
 
 
 
 
 
 
 
3f75218
a3e92cc
 
 
3f75218
 
 
8fefdb0
3f75218
58fbd56
3f75218
 
 
 
 
 
 
 
 
0f197e6
 
 
 
 
3f75218
 
 
 
 
b9bfa6b
3f75218
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
import argparse
from functools import partial
import cv2
import requests
import os
from io import BytesIO
from PIL import Image
import numpy as np
from pathlib import Path
import gradio as gr

import warnings

import torch

os.system("python setup.py build develop --user")
os.system("pip install packaging==21.3")
warnings.filterwarnings("ignore")


from groundingdino.models import build_model
from groundingdino.util.slconfig import SLConfig
from groundingdino.util.utils import clean_state_dict
from groundingdino.util.inference import annotate, load_image, predict
import groundingdino.datasets.transforms as T

from huggingface_hub import hf_hub_download



# Use this command for evaluate the GLIP-T model
config_file = "groundingdino/config/GroundingDINO_SwinT_OGC.py"
ckpt_repo_id = "ShilongLiu/GroundingDINO"
ckpt_filenmae = "groundingdino_swint_ogc.pth"


def load_model_hf(model_config_path, repo_id, filename, device='cpu'):
    args = SLConfig.fromfile(model_config_path) 
    model = build_model(args)
    args.device = device

    cache_file = hf_hub_download(repo_id=repo_id, filename=filename)
    checkpoint = torch.load(cache_file, map_location='cpu')
    log = model.load_state_dict(clean_state_dict(checkpoint['model']), strict=False)
    print("Model loaded from {} \n => {}".format(cache_file, log))
    _ = model.eval()
    return model    

def image_transform_grounding(init_image):
    transform = T.Compose([
        T.RandomResize([800], max_size=1333),
        T.ToTensor(),
        T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ])
    image, _ = transform(init_image, None) # 3, h, w
    return init_image, image

def image_transform_grounding_for_vis(init_image):
    transform = T.Compose([
        T.RandomResize([800], max_size=1333),
    ])
    image, _ = transform(init_image, None) # 3, h, w
    return image

model = load_model_hf(config_file, ckpt_repo_id, ckpt_filenmae)

def run_grounding(input_image, grounding_caption, box_threshold, text_threshold):
    init_image = input_image.convert("RGB")
    original_size = init_image.size

    _, image_tensor = image_transform_grounding(init_image)
    image_pil: Image = image_transform_grounding_for_vis(init_image)

    # run grounidng
    boxes, logits, phrases = predict(model, image_tensor, grounding_caption, box_threshold, text_threshold, device='cpu')
    annotated_frame = annotate(image_source=np.asarray(image_pil), boxes=boxes, logits=logits, phrases=phrases)
    image_with_box = Image.fromarray(cv2.cvtColor(annotated_frame, cv2.COLOR_BGR2RGB))


    return image_with_box

if __name__ == "__main__":
    
    parser = argparse.ArgumentParser("Grounding DINO demo", add_help=True)
    parser.add_argument("--debug", action="store_true", help="using debug mode")
    parser.add_argument("--share", action="store_true", help="share the app")
    args = parser.parse_args()
    css = """
  #mkd {
    height: 500px; 
    overflow: auto; 
    border: 1px solid #ccc; 
  }
"""
    block = gr.Blocks(css=css).queue()
    with block:
        gr.Markdown("<h1><center>Grounding DINO<h1><center>")
        gr.Markdown("<h3><center>Open-World Detection with <a href='https://github.com/IDEA-Research/GroundingDINO'>Grounding DINO</a><h3><center>")
        gr.Markdown("<h3><center>Note the model runs on CPU, so it may take a while to run the model.<h3><center>")

        with gr.Row():
            with gr.Column():
                input_image = gr.Image(type="pil")
                grounding_caption = gr.Textbox(label="Detection Prompt")
                run_button = gr.Button()
                with gr.Accordion("Advanced options", open=False):
                    box_threshold = gr.Slider(
                        label="Box Threshold", minimum=0.0, maximum=1.0, value=0.25, step=0.001
                    )
                    text_threshold = gr.Slider(
                        label="Text Threshold", minimum=0.0, maximum=1.0, value=0.25, step=0.001
                    )

            with gr.Column():
                # gallery = gr.outputs.Image(
                #     type="pil",
                #     # label="grounding results"
                # ).style(full_width=True, full_height=True)
                gallery = gr.Image(type="pil")
                # gallery = gr.Gallery(label="Generated images", show_label=False).style(
                #         grid=[1], height="auto", container=True, full_width=True, full_height=True)

        run_button.click(fn=run_grounding, inputs=[
                        input_image, grounding_caption, box_threshold, text_threshold], outputs=[gallery])
    block.launch(show_error=True)