File size: 6,157 Bytes
ab8156f
 
 
 
 
b291f6a
 
ab8156f
 
 
 
 
 
 
 
 
 
 
b291f6a
7b2c66b
 
 
ab8156f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
os.system("mim install 'mmengine>=0.7.0'")
os.system("mim install mmcv")
os.system("mim install 'mmdet>=3.0.0'") 
os.system("pip install -e .")


import numpy as np
import torch
from mmengine.config import Config
from mmengine.dataset import Compose
from mmengine.runner import Runner
from mmengine.runner.amp import autocast
from mmyolo.registry import RUNNERS
from torchvision.ops import nms
import supervision as sv
import PIL.Image
import cv2

import gradio as gr


TITLE = """
# YOLO-World-Seg

This is a demo of zero-shot object detection and instance segmentation using 
[YOLO-World](https://github.com/AILab-CVC/YOLO-World) 

Powered by [Supervision](https://github.com/roboflow/supervision).
"""

EXAMPLES = [
    ["https://media.roboflow.com/efficient-sam/corgi.jpg", "dog",0.5,0.5,0.5,100],
    ["https://media.roboflow.com/efficient-sam/horses.jpg", "horse",0.5,0.5,0.5,100],
    ["https://media.roboflow.com/efficient-sam/bears.jpg", "bear",0.5,0.5,0.5,100],
]

box_annotator = sv.BoxAnnotator()
label_annotator = sv.LabelAnnotator(text_position=sv.Position.CENTER)
mask_annotator = sv.MaskAnnotator(color_lookup=sv.ColorLookup.INDEX)

def load_runner():
    cfg = Config.fromfile(
    "./configs/segmentation/yolo_world_seg_l_dual_vlpan_2e-4_80e_8gpus_seghead_finetune_lvis.py"
    )
    cfg.work_dir = "."
    cfg.load_from = "yolo_world_seg_l_dual_vlpan_2e-4_80e_8gpus_seghead_finetune_lvis-5a642d30.pth"
    runner = Runner.from_cfg(cfg)
    runner.call_hook("before_run")
    runner.load_or_resume()
    pipeline = cfg.test_dataloader.dataset.pipeline
    runner.pipeline = Compose(pipeline)
    runner.model.eval()

def run_image(
        input_image,
        class_names="person,car,bus,truck",
        score_thr=0.05,
        iou_thr=0.5,
        nms_thr = 0.5,
        max_num_boxes=100,
        ):
    runner = load_runner()
    with open("input.jpeg", "wb") as f:
        f.write(input_image)
        
    class_names = [class_name.strip() for class_name in class_names.split(',')]
    
    texts = [[t.strip()] for t in class_names.split(",")] + [[" "]]
    data_info = runner.pipeline(dict(img_id=0, img_path="input.jpeg",
                                     texts=texts))

    data_batch = dict(
        inputs=data_info["inputs"].unsqueeze(0),
        data_samples=[data_info["data_samples"]],
    )

    with autocast(enabled=False), torch.no_grad():
        output = runner.model.test_step(data_batch)[0]
        runner.model.class_names = texts
        pred_instances = output.pred_instances

    keep_idxs = nms(pred_instances.bboxes, pred_instances.scores, iou_threshold=iou_thr)
    pred_instances = pred_instances[keep_idxs]
    pred_instances = pred_instances[pred_instances.scores.float() > score_thr]

    if len(pred_instances.scores) > max_num_boxes:
        indices = pred_instances.scores.float().topk(max_num_boxes)[1]
        pred_instances = pred_instances[indices]
    output.pred_instances = pred_instances
    result = pred_instances.cpu().numpy()
    detections = sv.Detections(
    xyxy=result['bboxes'],
    class_id=result['labels'],
    confidence=result['scores'],
    mask = result['masks']
    )
    detections = detections.with_nms(threshold=nms_thr)

    labels = [
        f"{class_id} {confidence:.3f}"
        for class_id, confidence
        in zip(detections.class_id, detections.confidence)
    ]
    
    svimage = box_annotator.annotate(input_image, detections)
    svimage = label_annotator.annotate(svimage, detections, labels)
    svimage = mask_annotator.annotate(svimage,detections)
    return svimage

confidence_threshold_component = gr.Slider(
    minimum=0,
    maximum=1.0,
    value=0.3,
    step=0.01,
    label="Confidence Threshold",
    info=(
        "The confidence threshold for the YOLO-World model. Lower the threshold to "
        "reduce false negatives, enhancing the model's sensitivity to detect "
        "sought-after objects. Conversely, increase the threshold to minimize false "
        "positives, preventing the model from identifying objects it shouldn't."
    ))

iou_threshold_component = gr.Slider(
    minimum=0,
    maximum=1.0,
    value=0.5,
    step=0.01,
    label="IoU Threshold",
    info=(
        "The Intersection over Union (IoU) threshold for non-maximum suppression. "
        "Decrease the value to lessen the occurrence of overlapping bounding boxes, "
        "making the detection process stricter. On the other hand, increase the value "
        "to allow more overlapping bounding boxes, accommodating a broader range of "
        "detections."
    ))

with gr.Blocks() as demo:
    gr.Markdown(TITLE)
    with gr.Accordion("Configuration", open=False):
        confidence_threshold_component.render()
        iou_threshold_component.render()
    with gr.Tab(label="Image"):
        with gr.Row():
            input_image_component = gr.Image(
                type='numpy',
                label='Input Image'
            )
            output_image_component = gr.Image(
                type='numpy',
                label='Output Image'
            )
        with gr.Row():
            image_categories_text_component = gr.Textbox(
                label='Categories',
                placeholder='comma separated list of categories',
                scale=7
            )
            image_submit_button_component = gr.Button(
                value='Submit',
                scale=1,
                variant='primary'
            )
        gr.Examples(
            fn=run_image,
            examples=EXAMPLES,
            inputs=[
                input_image_component,
                image_categories_text_component,
                confidence_threshold_component,
                iou_threshold_component,
            ],
            outputs=output_image_component
        )


    image_submit_button_component.click(
        fn=run_image,
        inputs=[
            input_image_component,
            image_categories_text_component,
            confidence_threshold_component,
            iou_threshold_component,
        ],
        outputs=output_image_component
    )
    


demo.launch(debug=False, show_error=True)