Update README.md
Browse files
README.md
CHANGED
@@ -4,9 +4,10 @@ license: mit
|
|
4 |
tags:
|
5 |
- pytorch
|
6 |
- image-segmentation
|
7 |
-
-
|
8 |
- glove
|
9 |
- baseball
|
|
|
10 |
- computer-vision
|
11 |
- custom-model
|
12 |
library_name: pytorch
|
@@ -15,8 +16,63 @@ datasets:
|
|
15 |
metrics:
|
16 |
- dice
|
17 |
- iou
|
18 |
-
inference:
|
19 |
widget: []
|
20 |
model-index:
|
21 |
- name: glove_labelling
|
22 |
results: []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
tags:
|
5 |
- pytorch
|
6 |
- image-segmentation
|
7 |
+
- sam2
|
8 |
- glove
|
9 |
- baseball
|
10 |
+
- sports-analytics
|
11 |
- computer-vision
|
12 |
- custom-model
|
13 |
library_name: pytorch
|
|
|
16 |
metrics:
|
17 |
- dice
|
18 |
- iou
|
19 |
+
inference: true
|
20 |
widget: []
|
21 |
model-index:
|
22 |
- name: glove_labelling
|
23 |
results: []
|
24 |
+
---
|
25 |
+
|
26 |
+
# Glove Labelling Model (SAM2 fine-tuned)
|
27 |
+
|
28 |
+
This repository contains a fine-tuned [SAM2](https://github.com/facebookresearch/sam2) hierarchical image segmentation model adapted for high-precision baseball glove segmentation.
|
29 |
+
|
30 |
+
### 💡 What it does
|
31 |
+
|
32 |
+
Given a frame from a pitching video, this model outputs per-pixel segmentations for:
|
33 |
+
|
34 |
+
- `glove_outline`
|
35 |
+
- `webbing`
|
36 |
+
- `thumb`
|
37 |
+
- `palm_pocket`
|
38 |
+
- `hand`
|
39 |
+
- `glove_exterior`
|
40 |
+
|
41 |
+
Trained on individual pitch frame sequences using COCO format masks.
|
42 |
+
|
43 |
+
---
|
44 |
+
|
45 |
+
### 🏗 Architecture
|
46 |
+
|
47 |
+
- Base Model: `SAM2Hierarchical`
|
48 |
+
- Framework: PyTorch
|
49 |
+
- Input shape: `[1, 3, 720, 1280]` RGB frame
|
50 |
+
- Output: Segmentation logits across 6 glove-related classes
|
51 |
+
|
52 |
+
---
|
53 |
+
|
54 |
+
### 🔧 Usage
|
55 |
+
|
56 |
+
To use the model for inference:
|
57 |
+
|
58 |
+
```python
|
59 |
+
import torch
|
60 |
+
from PIL import Image
|
61 |
+
import torchvision.transforms as T
|
62 |
+
|
63 |
+
model = torch.load("pytorch_model.bin", map_location="cpu")
|
64 |
+
model.eval()
|
65 |
+
|
66 |
+
transform = T.Compose([
|
67 |
+
T.Resize((720, 1280)),
|
68 |
+
T.ToTensor()
|
69 |
+
])
|
70 |
+
|
71 |
+
img = Image.open("example.jpg").convert("RGB")
|
72 |
+
x = transform(img).unsqueeze(0)
|
73 |
+
|
74 |
+
with torch.no_grad():
|
75 |
+
output = model(x)
|
76 |
+
|
77 |
+
# Convert logits to class labels
|
78 |
+
pred_mask = output.argmax(dim=1).squeeze().cpu().numpy()
|