test video processing on HF spaces
Browse files- app.py +61 -21
- requirements.txt +2 -1
app.py
CHANGED
@@ -1,5 +1,7 @@
|
|
1 |
-
|
|
|
2 |
|
|
|
3 |
import gradio as gr
|
4 |
import numpy as np
|
5 |
import supervision as sv
|
@@ -9,22 +11,16 @@ from rfdetr.detr import RFDETR
|
|
9 |
from rfdetr.util.coco_classes import COCO_CLASSES
|
10 |
|
11 |
from utils.image import calculate_resolution_wh
|
12 |
-
from utils.video import create_directory
|
|
|
|
|
13 |
|
14 |
MARKDOWN = """
|
15 |
# RF-DETR 🔥
|
16 |
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
</a>
|
21 |
-
<a href="https://blog.roboflow.com/rf-detr">
|
22 |
-
<img src="https://raw.githubusercontent.com/roboflow-ai/notebooks/main/assets/badges/roboflow-blogpost.svg" alt="roboflow" style="display:inline-block;">
|
23 |
-
</a>
|
24 |
-
<a href="https://github.com/roboflow/rf-detr">
|
25 |
-
<img src="https://badges.aleen42.com/src/github.svg" alt="roboflow" style="display:inline-block;">
|
26 |
-
</a>
|
27 |
-
</div>
|
28 |
|
29 |
RF-DETR is a real-time, transformer-based object detection model architecture developed
|
30 |
by [Roboflow](https://roboflow.com/) and released under the Apache 2.0 license.
|
@@ -41,12 +37,18 @@ COLOR = sv.ColorPalette.from_hex([
|
|
41 |
"#9999ff", "#3399ff", "#66ffff", "#33ff99", "#66ff66", "#99ff00"
|
42 |
])
|
43 |
|
|
|
44 |
VIDEO_SCALE_FACTOR = 0.5
|
45 |
VIDEO_TARGET_DIRECTORY = "tmp"
|
|
|
46 |
create_directory(directory_path=VIDEO_TARGET_DIRECTORY)
|
47 |
|
48 |
|
49 |
-
def detect_and_annotate(
|
|
|
|
|
|
|
|
|
50 |
detections = model.predict(image, threshold=confidence)
|
51 |
|
52 |
resolution_wh = calculate_resolution_wh(image)
|
@@ -73,16 +75,54 @@ def detect_and_annotate(model: RFDETR, image: Union[Image.Image, np.ndarray], co
|
|
73 |
return annotated_image
|
74 |
|
75 |
|
76 |
-
def
|
77 |
-
|
78 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
79 |
return detect_and_annotate(model=model, image=input_image, confidence=confidence)
|
80 |
|
81 |
|
82 |
-
def video_processing_inference(
|
83 |
-
|
84 |
-
|
85 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
86 |
|
87 |
with gr.Blocks() as demo:
|
88 |
gr.Markdown(MARKDOWN)
|
|
|
1 |
+
import os
|
2 |
+
from typing import TypeVar
|
3 |
|
4 |
+
from tqdm import tqdm
|
5 |
import gradio as gr
|
6 |
import numpy as np
|
7 |
import supervision as sv
|
|
|
11 |
from rfdetr.util.coco_classes import COCO_CLASSES
|
12 |
|
13 |
from utils.image import calculate_resolution_wh
|
14 |
+
from utils.video import create_directory, generate_unique_name
|
15 |
+
|
16 |
+
ImageType = TypeVar("ImageType", Image.Image, np.ndarray)
|
17 |
|
18 |
MARKDOWN = """
|
19 |
# RF-DETR 🔥
|
20 |
|
21 |
+
[`[code]`](https://github.com/roboflow/rf-detr)
|
22 |
+
[`[blog]`](https://blog.roboflow.com/rf-detr)
|
23 |
+
[`[notebook]`](https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/how-to-finetune-rf-detr-on-detection-dataset.ipynb)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
|
25 |
RF-DETR is a real-time, transformer-based object detection model architecture developed
|
26 |
by [Roboflow](https://roboflow.com/) and released under the Apache 2.0 license.
|
|
|
37 |
"#9999ff", "#3399ff", "#66ffff", "#33ff99", "#66ff66", "#99ff00"
|
38 |
])
|
39 |
|
40 |
+
MAX_VIDEO_LENGTH_SECONDS = 2
|
41 |
VIDEO_SCALE_FACTOR = 0.5
|
42 |
VIDEO_TARGET_DIRECTORY = "tmp"
|
43 |
+
|
44 |
create_directory(directory_path=VIDEO_TARGET_DIRECTORY)
|
45 |
|
46 |
|
47 |
+
def detect_and_annotate(
|
48 |
+
model: RFDETR,
|
49 |
+
image: ImageType,
|
50 |
+
confidence: float
|
51 |
+
) -> ImageType:
|
52 |
detections = model.predict(image, threshold=confidence)
|
53 |
|
54 |
resolution_wh = calculate_resolution_wh(image)
|
|
|
75 |
return annotated_image
|
76 |
|
77 |
|
78 |
+
def load_model(resolution: int, checkpoint: str) -> RFDETR:
|
79 |
+
if checkpoint == "base":
|
80 |
+
return RFDETRBase(resolution=resolution)
|
81 |
+
elif checkpoint == "large":
|
82 |
+
return RFDETRLarge(resolution=resolution)
|
83 |
+
raise TypeError("Checkpoint must be a base or large.")
|
84 |
+
|
85 |
+
|
86 |
+
def image_processing_inference(
|
87 |
+
input_image: Image.Image,
|
88 |
+
confidence: float,
|
89 |
+
resolution: int,
|
90 |
+
checkpoint: str
|
91 |
+
):
|
92 |
+
model = load_model(resolution=resolution, checkpoint=checkpoint)
|
93 |
return detect_and_annotate(model=model, image=input_image, confidence=confidence)
|
94 |
|
95 |
|
96 |
+
def video_processing_inference(
|
97 |
+
input_video: str,
|
98 |
+
confidence: float,
|
99 |
+
resolution: int,
|
100 |
+
checkpoint: str,
|
101 |
+
progress=gr.Progress(track_tqdm=True)
|
102 |
+
):
|
103 |
+
model = load_model(resolution=resolution, checkpoint=checkpoint)
|
104 |
+
|
105 |
+
name = generate_unique_name()
|
106 |
+
output_video = os.path.join(VIDEO_TARGET_DIRECTORY, f"{name}.mp4")
|
107 |
+
|
108 |
+
video_info = sv.VideoInfo.from_video_path(input_video)
|
109 |
+
video_info.width = int(video_info.width * VIDEO_SCALE_FACTOR)
|
110 |
+
video_info.height = int(video_info.height * VIDEO_SCALE_FACTOR)
|
111 |
+
|
112 |
+
total = min(video_info.total_frames, video_info.fps * MAX_VIDEO_LENGTH_SECONDS)
|
113 |
+
frames_generator = sv.get_video_frames_generator(input_video, end=total)
|
114 |
+
|
115 |
+
with sv.VideoSink(output_video, video_info=video_info) as sink:
|
116 |
+
for frame in tqdm(frames_generator, total=total):
|
117 |
+
frame = sv.scale_image(frame, VIDEO_SCALE_FACTOR)
|
118 |
+
annotated_frame = detect_and_annotate(
|
119 |
+
model=model,
|
120 |
+
image=frame,
|
121 |
+
confidence=confidence
|
122 |
+
)
|
123 |
+
sink.write_frame(annotated_frame)
|
124 |
+
|
125 |
+
return output_video
|
126 |
|
127 |
with gr.Blocks() as demo:
|
128 |
gr.Markdown(MARKDOWN)
|
requirements.txt
CHANGED
@@ -1,3 +1,4 @@
|
|
1 |
gradio
|
2 |
spaces
|
3 |
-
rfdetr
|
|
|
|
1 |
gradio
|
2 |
spaces
|
3 |
+
rfdetr
|
4 |
+
tqdm
|