nlateralus commited on
Commit
b484048
·
verified ·
1 Parent(s): 9c543a2

Model 20261119-231345, validation water IoU 0.996

Browse files
Files changed (3) hide show
  1. README.md +48 -0
  2. config.json +19 -0
  3. model.safetensors +3 -0
README.md ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ pipeline_tag: image-segmentation
4
+ tags:
5
+ - pytorch
6
+ - u-net
7
+ - swimming-pool
8
+ - semantic-segmentation
9
+ ---
10
+
11
+ # pool-segmentation: a swimming pool's water
12
+
13
+ A small U-Net (resnet18 encoder) that finds a swimming pool's **water** in video frames: a class for every
14
+ pixel, 0 other, 1 water (swimmers in the water, splashes and lane ropes count as water). About 25 ms per frame on a
15
+ MacBook. Made for swimming video analysis (dropping people outside the pool; later, finding the lanes).
16
+
17
+ Code, training and labeling tools: [github.com/Neptune-Trojans/pool-segmentation](https://github.com/Neptune-Trojans/pool-segmentation) (`pip install pool-segmentation`).
18
+
19
+ ## Use
20
+
21
+ ```python
22
+ from pool_segmentation import PoolModel, find_model
23
+
24
+ model = PoolModel.load(find_model()) # your newest training run in data/models/pool/, else this model
25
+ classes = model.predict(frame) # BGR frame -> class per pixel: 0 other, 1 water
26
+ water = model.water(frame) # the water's outline: one polygon [(x, y), ...] per piece of water
27
+ ```
28
+
29
+ The package downloads this model the first time it is needed and keeps it in `~/.cache/huggingface`; when online, it
30
+ gets a newer one when there is one. A given version: `find_model("hub:20261119-231345")` (every version is tagged
31
+ with its training run).
32
+
33
+ ## This version
34
+
35
+ | | |
36
+ |---|---|
37
+ | Training run | `20261119-231345` |
38
+ | Classes | 0 other, 1 water |
39
+ | Input | frames scaled to 1024×576 (the long side; shape kept) |
40
+ | Training | 89 hand-checked frames from 9 videos, 40 epochs |
41
+ | Validation | 19 frames (every 6th of each video): water IoU mean 0.996, worst 0.990 |
42
+
43
+ ## Limits
44
+
45
+ Trained on frames filmed at a few pools, from the deck. Pools that look very different (another camera angle, light,
46
+ water color) may need some frames of their own: the package's labeling tools pick frames from your videos, let this
47
+ model label them, and you check them in the browser before training again. The training frames are not published:
48
+ they show real people.
config.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run": "20261119-231345",
3
+ "classes": [
4
+ "other",
5
+ "water"
6
+ ],
7
+ "encoder": "resnet18",
8
+ "size": [
9
+ 1024,
10
+ 576
11
+ ],
12
+ "created": "2026-11-19T23:19:56",
13
+ "epochs": 40,
14
+ "train_frames": 89,
15
+ "val_frames": 19,
16
+ "videos": 9,
17
+ "val_iou_mean": 0.9956,
18
+ "val_iou_min": 0.9901
19
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:01654196f4f96fe4b3e2848690fa4248f850d0e49506101756171a148239801e
3
+ size 57375560