Spaces:
Running
Running
feat: mvp of the space
Browse files- .gitattributes +1 -0
- .gitignore +1 -0
- .python-version +1 -0
- README.md +4 -3
- app.py +169 -0
- data/model/weights.pt +3 -0
- data/videos/video1-clip.mp4 +3 -0
- requirements.txt +2 -0
.gitattributes
CHANGED
@@ -32,4 +32,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
32 |
*.xz filter=lfs diff=lfs merge=lfs -text
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
|
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
32 |
*.xz filter=lfs diff=lfs merge=lfs -text
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
+
*.mp4 filter=lfs diff=lfs merge=lfs -text
|
36 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
.gitignore
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
runs/
|
.python-version
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
3.10.12
|
README.md
CHANGED
@@ -1,13 +1,14 @@
|
|
1 |
---
|
2 |
title: Salmon Vision
|
3 |
-
emoji:
|
4 |
colorFrom: red
|
5 |
-
colorTo:
|
|
|
6 |
sdk: gradio
|
7 |
sdk_version: 5.5.0
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
-
short_description:
|
11 |
---
|
12 |
|
13 |
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
|
1 |
---
|
2 |
title: Salmon Vision
|
3 |
+
emoji: π
|
4 |
colorFrom: red
|
5 |
+
colorTo: blue
|
6 |
+
python_version: 3.10.12
|
7 |
sdk: gradio
|
8 |
sdk_version: 5.5.0
|
9 |
app_file: app.py
|
10 |
pinned: false
|
11 |
+
short_description: Wild salmon migration monitoring
|
12 |
---
|
13 |
|
14 |
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
app.py
ADDED
@@ -0,0 +1,169 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
Gradio app to showcase the pyronear model for early forest fire detection.
|
3 |
+
"""
|
4 |
+
|
5 |
+
from collections import Counter
|
6 |
+
from pathlib import Path
|
7 |
+
from typing import Any, Tuple
|
8 |
+
|
9 |
+
import gradio as gr
|
10 |
+
import numpy as np
|
11 |
+
from ultralytics import YOLO
|
12 |
+
|
13 |
+
|
14 |
+
def bgr_to_rgb(a: np.ndarray) -> np.ndarray:
|
15 |
+
"""
|
16 |
+
Turn a BGR numpy array into a RGB numpy array when the array `a` represents
|
17 |
+
an image.
|
18 |
+
"""
|
19 |
+
return a[:, :, ::-1]
|
20 |
+
|
21 |
+
|
22 |
+
def analyze_predictions(yolo_predictions) -> dict[str, Any]:
|
23 |
+
"""
|
24 |
+
Analyze the raw `yolo_predictions` and outputs a dict containg information.
|
25 |
+
|
26 |
+
Args:
|
27 |
+
yolo_predictions: result of calling model.track() on a video
|
28 |
+
|
29 |
+
Returns:
|
30 |
+
counts (int): number of distinct identifiers.
|
31 |
+
ids (set[int]): all the assigned identifiers.
|
32 |
+
detected_species (dict[int, int]): mapping from identifier to instance class
|
33 |
+
names (list[str]): the class names used by the model
|
34 |
+
"""
|
35 |
+
if len(yolo_predictions) == 0:
|
36 |
+
return {
|
37 |
+
"counts": 0,
|
38 |
+
"ids": set(),
|
39 |
+
"detected_species": {},
|
40 |
+
"names": None,
|
41 |
+
}
|
42 |
+
else:
|
43 |
+
names = yolo_predictions[0].names
|
44 |
+
ids = set()
|
45 |
+
for prediction in yolo_predictions:
|
46 |
+
if prediction.boxes.id:
|
47 |
+
for id in prediction.boxes.id.numpy().astype("int"):
|
48 |
+
ids.add(id.item())
|
49 |
+
detected_species = {}
|
50 |
+
for id in ids:
|
51 |
+
counter = Counter()
|
52 |
+
for prediction in yolo_predictions:
|
53 |
+
if prediction.boxes.id:
|
54 |
+
for idd, klass in zip(
|
55 |
+
prediction.boxes.id.numpy().astype("int"),
|
56 |
+
prediction.boxes.cls.numpy().astype("int"),
|
57 |
+
):
|
58 |
+
if idd.item() == id:
|
59 |
+
counter[klass.item()] += 1
|
60 |
+
selected_class = counter.most_common(1)[0][0]
|
61 |
+
detected_species[id] = selected_class
|
62 |
+
return {
|
63 |
+
"counts": len(ids),
|
64 |
+
"ids": ids,
|
65 |
+
"detected_species": detected_species,
|
66 |
+
"names": names,
|
67 |
+
}
|
68 |
+
|
69 |
+
|
70 |
+
def prediction_to_str(yolo_predictions) -> str:
|
71 |
+
"""
|
72 |
+
Turn the yolo_predictions into a human friendly string.
|
73 |
+
"""
|
74 |
+
if len(yolo_predictions) == 0:
|
75 |
+
return "No prediction"
|
76 |
+
else:
|
77 |
+
result = analyze_predictions(yolo_predictions=yolo_predictions)
|
78 |
+
names = result["names"]
|
79 |
+
detected_species = result["detected_species"]
|
80 |
+
ids = result["ids"]
|
81 |
+
summary_str = "\n".join(
|
82 |
+
[
|
83 |
+
f"- The fish with id {id} is a {names.get(klass, 'Unknown')}"
|
84 |
+
for id, klass in detected_species.items()
|
85 |
+
]
|
86 |
+
)
|
87 |
+
print(summary_str)
|
88 |
+
return f"Detected {len(ids)} salmons in the video clip with ids {ids}:\n{summary_str}"
|
89 |
+
|
90 |
+
|
91 |
+
def predict(model: YOLO, video_filepath: Path) -> Tuple[Path, str]:
|
92 |
+
"""
|
93 |
+
Main interface function that runs the model on the provided pil_image and
|
94 |
+
returns the exepected tuple to populate the gradio interface.
|
95 |
+
|
96 |
+
Args:
|
97 |
+
model (YOLO): Loaded ultralytics YOLO model.
|
98 |
+
pil_image (PIL): image to run inference on.
|
99 |
+
|
100 |
+
Returns:
|
101 |
+
pil_image_with_prediction (PIL): image with prediction from the model.
|
102 |
+
raw_prediction_str (str): string representing the raw prediction from the
|
103 |
+
model.
|
104 |
+
"""
|
105 |
+
project = "runs/track/"
|
106 |
+
name = video_filepath.stem
|
107 |
+
predictions = model.track(
|
108 |
+
source=video_filepath,
|
109 |
+
save=True,
|
110 |
+
tracker="bytetrack.yaml",
|
111 |
+
exist_ok=True,
|
112 |
+
project=project,
|
113 |
+
name=name,
|
114 |
+
)
|
115 |
+
filepath_video_prediction = Path(f"{project}/{name}/{name}.avi")
|
116 |
+
raw_prediction_str = prediction_to_str(yolo_predictions=predictions)
|
117 |
+
return (filepath_video_prediction, raw_prediction_str)
|
118 |
+
|
119 |
+
|
120 |
+
def examples(dir_examples: Path) -> list[Path]:
|
121 |
+
"""
|
122 |
+
List the images from the dir_examples directory.
|
123 |
+
|
124 |
+
Returns:
|
125 |
+
filepaths (list[Path]): list of image filepaths.
|
126 |
+
"""
|
127 |
+
return list(dir_examples.glob("*.mp4"))
|
128 |
+
|
129 |
+
|
130 |
+
def load_model(filepath_weights: Path) -> YOLO:
|
131 |
+
"""
|
132 |
+
Load the YOLO model given the filepath_weights.
|
133 |
+
"""
|
134 |
+
return YOLO(filepath_weights)
|
135 |
+
|
136 |
+
|
137 |
+
# Main Gradio interface
|
138 |
+
|
139 |
+
MODEL_FILEPATH_WEIGHTS = Path("data/model/weights.pt")
|
140 |
+
DIR_EXAMPLES = Path("data/videos/")
|
141 |
+
DEFAULT_IMAGE_INDEX = 0
|
142 |
+
|
143 |
+
with gr.Blocks() as demo:
|
144 |
+
model = load_model(MODEL_FILEPATH_WEIGHTS)
|
145 |
+
videos_filepaths = examples(dir_examples=DIR_EXAMPLES)
|
146 |
+
print(f"videos_filepaths: {videos_filepaths}")
|
147 |
+
default_value_input = videos_filepaths[DEFAULT_IMAGE_INDEX]
|
148 |
+
input = gr.Video(
|
149 |
+
value=default_value_input,
|
150 |
+
format="mp4",
|
151 |
+
label="input video",
|
152 |
+
sources=["upload"],
|
153 |
+
)
|
154 |
+
output_video = gr.Video(format="mp4", label="model prediction")
|
155 |
+
output_raw = gr.Text(label="raw prediction")
|
156 |
+
|
157 |
+
fn = lambda video_filepath: predict(
|
158 |
+
model=model, video_filepath=Path(video_filepath)
|
159 |
+
)
|
160 |
+
gr.Interface(
|
161 |
+
title="ML model for wild salmon migration monitoring π",
|
162 |
+
fn=fn,
|
163 |
+
inputs=input,
|
164 |
+
outputs=[output_video, output_raw],
|
165 |
+
examples=videos_filepaths,
|
166 |
+
flagging_mode="never",
|
167 |
+
)
|
168 |
+
|
169 |
+
demo.launch()
|
data/model/weights.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:33d5e39e94c3f94badb476743ec9773f5df09b3f2755379f43b3a594fa755bd2
|
3 |
+
size 6239129
|
data/videos/video1-clip.mp4
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:3fb22eeeb3be60b9cb2a65d54a2d6c5379e9e65c47d7f2ce88eab0986f069167
|
3 |
+
size 2966504
|
requirements.txt
ADDED
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
1 |
+
ultralytics==8.3.*
|
2 |
+
gradio==5.4.*
|