File size: 2,741 Bytes
e7a52dc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
adb211e
e7a52dc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: apache-2.0
library_name: pytorch
pipeline_tag: image-classification
tags:
  - image-classification
  - vision
  - cats
  - pytorch
---

# stackmeo 🐱

stackmeo is a small image-classification model that recognizes cat breeds from a photo.
The name comes from "mèo", Vietnamese for cat.

Give it one picture of a cat and it returns the most likely breeds, each with a confidence score.
It's kept lightweight so it's easy to drop into a hobby project, a demo, or an app.

## What it recognizes

stackmeo knows 10 breeds:

- **Maine Coon**: a large, friendly cat known for its bushy tail, tufted ears, and playful personality.
- **Ragdoll**: a calm, affectionate cat that tends to go limp when you hold it.
- **British Shorthair**: a sturdy, quiet, and easygoing cat with a dense coat and round face.
- **Siamese**: a slender, vocal cat with striking blue eyes and sharp color points.
- **Persian**: a fluffy, quiet "couch potato" with a flat face that needs daily grooming.
- **Sphynx**: a hairless, high-energy cat with big ears and warm, suede-like skin.
- **Bengal**: an active, athletic cat with a spotted coat that looks like a wild leopard.
- **Scottish Fold**: a sweet-tempered cat known for its folded-over ears and round eyes.
- **Russian Blue**: a graceful cat with a shimmering silver-gray coat and a gentle, quiet nature.
- **Abyssinian**: a lean, curious, and playful cat with a ticked, multi-colored coat.

## Quick start

```bash
pip install torch torchvision pillow huggingface_hub
```

```python
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from torchvision import transforms

path = hf_hub_download("hgrrg/stackmeo", "test.pt")
model = torch.jit.load(path, map_location="cpu").eval()

LABELS = [
    "Maine Coon", "Ragdoll", "British Shorthair", "Siamese", "Persian",
    "Sphynx", "Bengal", "Scottish Fold", "Russian Blue", "Abyssinian",
]

preprocess = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
])

img = preprocess(Image.open("my_cat.jpg").convert("RGB")).unsqueeze(0)
with torch.no_grad():
    probs = model(img).softmax(dim=1)[0]

top = probs.topk(3)
for score, idx in zip(top.values, top.indices):
    print(f"{LABELS[idx]}: {score:.1%}")
```

This prints the 3 most likely breeds for `my_cat.jpg` with a confidence score for each.

## Limitations

- It only knows cats. Show it a dog or a teapot and it will still answer with a cat breed,
  because it can't say "this isn't a cat".
- Mixed-breed cats, and breeds that aren't in the list above, get matched to the closest known breed.
- It works best on clear photos where the cat fills most of the frame.