File size: 8,699 Bytes
75be6c3
 
1015457
41975a2
094752c
1015457
094752c
1015457
377e9f4
1015457
094752c
 
1015457
75be6c3
 
 
e317241
094752c
 
 
75be6c3
 
 
094752c
 
 
 
 
8a7385d
e317241
 
 
377e9f4
8a7385d
 
 
 
377e9f4
094752c
 
 
 
 
8a7385d
e317241
 
75be6c3
e317241
094752c
 
75be6c3
 
 
 
 
e317241
094752c
1015457
 
 
094752c
 
 
 
 
8a7385d
094752c
 
 
 
 
 
 
 
 
 
 
 
 
1015457
75be6c3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1015457
 
 
75be6c3
 
 
 
6057bcc
 
75be6c3
 
 
 
 
 
 
 
 
 
 
 
 
 
41975a2
75be6c3
 
 
 
 
8a7385d
75be6c3
 
 
1015457
094752c
 
1015457
 
 
 
094752c
 
 
 
1015457
 
 
 
 
e317241
1015457
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
377e9f4
 
 
 
1015457
 
 
 
 
8a7385d
1015457
 
 
 
 
 
 
8a7385d
 
1015457
 
 
 
 
 
 
 
 
 
8a7385d
1015457
 
 
 
 
094752c
 
1015457
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8a7385d
 
 
 
 
 
 
 
 
 
 
 
 
1015457
 
 
 
 
 
 
 
 
 
094752c
 
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
import os
from typing import TypeVar

from tqdm import tqdm
import gradio as gr
import numpy as np
import supervision as sv
from PIL import Image
from rfdetr import RFDETRBase, RFDETRLarge
from rfdetr.detr import RFDETR
from rfdetr.util.coco_classes import COCO_CLASSES

from utils.image import calculate_resolution_wh
from utils.video import create_directory, generate_unique_name

ImageType = TypeVar("ImageType", Image.Image, np.ndarray)

MARKDOWN = """
# RF-DETR 🔥

[`[code]`](https://github.com/roboflow/rf-detr) 
[`[blog]`](https://blog.roboflow.com/rf-detr) 
[`[notebook]`](https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/how-to-finetune-rf-detr-on-detection-dataset.ipynb)

RF-DETR is a real-time, transformer-based object detection model architecture developed 
by [Roboflow](https://roboflow.com/) and released under the Apache 2.0 license.
"""

IMAGE_PROCESSING_EXAMPLES = [
    ['https://media.roboflow.com/supervision/image-examples/people-walking.png', 0.3, 728, "large"],
    ['https://media.roboflow.com/supervision/image-examples/vehicles.png', 0.3, 728, "large"],
    ['https://media.roboflow.com/notebooks/examples/dog-2.jpeg', 0.5, 560, "base"],
]
VIDEO_PROCESSING_EXAMPLES = [
    ["videos/people-walking.mp4", 0.3, 728, "large"],
    ["videos/vehicles.mp4", 0.3, 728, "large"],
]

COLOR = sv.ColorPalette.from_hex([
    "#ffff00", "#ff9b00", "#ff8080", "#ff66b2", "#ff66ff", "#b266ff",
    "#9999ff", "#3399ff", "#66ffff", "#33ff99", "#66ff66", "#99ff00"
])

MAX_VIDEO_LENGTH_SECONDS = 5
VIDEO_SCALE_FACTOR = 0.5
VIDEO_TARGET_DIRECTORY = "tmp"

create_directory(directory_path=VIDEO_TARGET_DIRECTORY)


def detect_and_annotate(
        model: RFDETR,
        image: ImageType,
        confidence: float
) -> ImageType:
    detections = model.predict(image, threshold=confidence)

    resolution_wh = calculate_resolution_wh(image)
    text_scale = sv.calculate_optimal_text_scale(resolution_wh=resolution_wh) - 0.2
    thickness = sv.calculate_optimal_line_thickness(resolution_wh=resolution_wh)

    bbox_annotator = sv.BoxAnnotator(color=COLOR, thickness=thickness)
    label_annotator = sv.LabelAnnotator(
        color=COLOR,
        text_color=sv.Color.BLACK,
        text_scale=text_scale
    )

    labels = [
        f"{COCO_CLASSES[class_id]} {confidence:.2f}"
        for class_id, confidence
        in zip(detections.class_id, detections.confidence)
    ]

    annotated_image = image.copy()
    annotated_image = bbox_annotator.annotate(annotated_image, detections)
    annotated_image = label_annotator.annotate(annotated_image, detections, labels)
    return annotated_image


def load_model(resolution: int, checkpoint: str) -> RFDETR:
    if checkpoint == "base":
        return RFDETRBase(resolution=resolution)
    elif checkpoint == "large":
        return RFDETRLarge(resolution=resolution)
    raise TypeError("Checkpoint must be a base or large.")


def image_processing_inference(
        input_image: Image.Image,
        confidence: float,
        resolution: int,
        checkpoint: str
):
    model = load_model(resolution=resolution, checkpoint=checkpoint)
    return detect_and_annotate(model=model, image=input_image, confidence=confidence)


def video_processing_inference(
        input_video: str,
        confidence: float,
        resolution: int,
        checkpoint: str,
        progress=gr.Progress(track_tqdm=True)
):
    model = load_model(resolution=resolution, checkpoint=checkpoint)

    name = generate_unique_name()
    output_video = os.path.join(VIDEO_TARGET_DIRECTORY, f"{name}.mp4")

    video_info = sv.VideoInfo.from_video_path(input_video)
    video_info.width = int(video_info.width * VIDEO_SCALE_FACTOR)
    video_info.height = int(video_info.height * VIDEO_SCALE_FACTOR)

    total = min(video_info.total_frames, video_info.fps * MAX_VIDEO_LENGTH_SECONDS)
    frames_generator = sv.get_video_frames_generator(input_video, end=total)

    with sv.VideoSink(output_video, video_info=video_info) as sink:
        for frame in tqdm(frames_generator, total=total):
            annotated_frame = detect_and_annotate(
                model=model,
                image=frame,
                confidence=confidence
            )
            annotated_frame = sv.scale_image(annotated_frame, VIDEO_SCALE_FACTOR)
            sink.write_frame(annotated_frame)

    return output_video

with gr.Blocks() as demo:
    gr.Markdown(MARKDOWN)
    with gr.Tab("Image"):
        with gr.Row():
            image_processing_input_image = gr.Image(
                label="Upload image",
                image_mode='RGB',
                type='pil',
                height=600
            )
            image_processing_output_image = gr.Image(
                label="Output image",
                image_mode='RGB',
                type='pil',
                height=600
            )
        with gr.Row():
            with gr.Column():
                image_processing_confidence_slider = gr.Slider(
                    label="Confidence",
                    minimum=0.0,
                    maximum=1.0,
                    step=0.05,
                    value=0.5,
                )
                image_processing_resolution_slider = gr.Slider(
                    label="Inference resolution",
                    minimum=560,
                    maximum=1120,
                    step=56,
                    value=728,
                )
                image_processing_checkpoint_dropdown = gr.Dropdown(
                    label="Checkpoint",
                    choices=["base", "large"],
                    value="base"
                )
            with gr.Column():
                image_processing_submit_button = gr.Button("Submit", value="primary")

        gr.Examples(
            fn=image_processing_inference,
            examples=IMAGE_PROCESSING_EXAMPLES,
            inputs=[
                image_processing_input_image,
                image_processing_confidence_slider,
                image_processing_resolution_slider,
                image_processing_checkpoint_dropdown
            ],
            outputs=image_processing_output_image,
            cache_examples=True,
            run_on_click=True
        )

        image_processing_submit_button.click(
            image_processing_inference,
            inputs=[
                image_processing_input_image,
                image_processing_confidence_slider,
                image_processing_resolution_slider,
                image_processing_checkpoint_dropdown
            ],
            outputs=image_processing_output_image,
        )
    with gr.Tab("Video"):
        with gr.Row():
            video_processing_input_video = gr.Video(
                label='Upload video',
                height=600
            )
            video_processing_output_video = gr.Video(
                label='Output video',
                height=600
            )
        with gr.Row():
            with gr.Column():
                video_processing_confidence_slider = gr.Slider(
                    label="Confidence",
                    minimum=0.0,
                    maximum=1.0,
                    step=0.05,
                    value=0.5,
                )
                video_processing_resolution_slider = gr.Slider(
                    label="Inference resolution",
                    minimum=560,
                    maximum=1120,
                    step=56,
                    value=728,
                )
                video_processing_checkpoint_dropdown = gr.Dropdown(
                    label="Checkpoint",
                    choices=["base", "large"],
                    value="base"
                )
            with gr.Column():
                video_processing_submit_button = gr.Button("Submit", value="primary")

        gr.Examples(
            fn=video_processing_inference,
            examples=VIDEO_PROCESSING_EXAMPLES,
            inputs=[
                video_processing_input_video,
                video_processing_confidence_slider,
                video_processing_resolution_slider,
                video_processing_checkpoint_dropdown
            ],
            outputs=video_processing_output_video,
            run_on_click=True
        )

        video_processing_submit_button.click(
            video_processing_inference,
            inputs=[
                video_processing_input_video,
                video_processing_confidence_slider,
                video_processing_resolution_slider,
                video_processing_checkpoint_dropdown
            ],
            outputs=video_processing_output_video
        )

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