timm
/

Image Classification
timm
PyTorch
Safetensors
File size: 8,253 Bytes
5c13342
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8d244cb
 
5c13342
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
524645a
 
 
8d244cb
 
5221dd8
524645a
 
 
5221dd8
524645a
 
 
 
 
3ae6b70
524645a
5c13342
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
---
tags:
- image-classification
- timm
library_name: timm
license: mit
datasets:
- imagenet-1k
---
# Model card for vit_medium_patch16_rope_reg1_gap_256.sbb_in1k

A Vision Transformer (ViT) image classification model. This is a `timm` specific variation of the architecture with rotary position embeddings (ROPE), registers, global average pooling.

There are a number of models in the lower end of model scales that originate in `timm`:

| variant | width | mlp width (mult) | heads | depth | timm orig |
| ------- | ----- | ---------------- | ----- | ----- | ---- |
| tiny | 192 | 768 (4) | 3 | 12 | n |
| wee | 256 | 1280 (5) | 4 | 14 | y |
| pwee | 256 | 1280 (5) | 4 | 16 (parallel) | y |
| small | 384 | 1536 (4) | 6 | 12 | n |
| little | 320 | 1792 (5.6) | 5 | 14 | y |
| medium | 512 | 2048 (4) | 8 | 12 | y |
| mediumd | 512 | 2048 (4) | 8 | 20 | y |
| betwixt | 640 | 2560 (4) | 10 | 12 | y |
| base | 768 | 3072 (4) | 12 | 12 | n |

Trained on ImageNet-1k in `timm` using recipe template described below.

Recipe details:
 * Searching for better baselines. Influced by Swin/DeiT/DeiT-III but w/ increased weight decay, moderate (in12k) to high (in1k) augmentation. Layer-decay used for fine-tune. Some runs used BCE and/or NAdamW instead of AdamW.
 * See [train_hparams.yaml](./train_hparams.yaml) for specifics of each model.


## Model Details
- **Model Type:** Image classification / feature backbone
- **Model Stats:**
  - Params (M): 38.7
  - GMACs: 9.8
  - Activations (M): 14.3
  - Image size: 256 x 256
- **Papers:**
  - EVA-02: A Visual Representation for Neon Genesis: https://arxiv.org/abs/2303.11331
  - Vision Transformers Need Registers: https://arxiv.org/abs/2309.16588
  - An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale: https://arxiv.org/abs/2010.11929v2
- **Dataset:** ImageNet-1k
- **Original:** https://github.com/huggingface/pytorch-image-models

## Model Usage
### Image Classification
```python
from urllib.request import urlopen
from PIL import Image
import timm

img = Image.open(urlopen(
    'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))

model = timm.create_model('vit_medium_patch16_rope_reg1_gap_256.sbb_in1k', pretrained=True)
model = model.eval()

# get model specific transforms (normalization, resize)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)

output = model(transforms(img).unsqueeze(0))  # unsqueeze single image into batch of 1

top5_probabilities, top5_class_indices = torch.topk(output.softmax(dim=1) * 100, k=5)
```

### Feature Map Extraction
```python
from urllib.request import urlopen
from PIL import Image
import timm

img = Image.open(urlopen(
    'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))

model = timm.create_model(
    'vit_medium_patch16_rope_reg1_gap_256.sbb_in1k',
    pretrained=True,
    features_only=True,
)
model = model.eval()

# get model specific transforms (normalization, resize)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)

output = model(transforms(img).unsqueeze(0))  # unsqueeze single image into batch of 1

for o in output:
    # print shape of each feature map in output
    # e.g.:
    #  torch.Size([1, 512, 16, 16])
    #  torch.Size([1, 512, 16, 16])
    #  torch.Size([1, 512, 16, 16])

    print(o.shape)
```

### Image Embeddings
```python
from urllib.request import urlopen
from PIL import Image
import timm

img = Image.open(urlopen(
    'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))

model = timm.create_model(
    'vit_medium_patch16_rope_reg1_gap_256.sbb_in1k',
    pretrained=True,
    num_classes=0,  # remove classifier nn.Linear
)
model = model.eval()

# get model specific transforms (normalization, resize)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)

output = model(transforms(img).unsqueeze(0))  # output is (batch_size, num_features) shaped tensor

# or equivalently (without needing to set num_classes=0)

output = model.forward_features(transforms(img).unsqueeze(0))
# output is unpooled, a (1, 257, 512) shaped tensor

output = model.forward_head(output, pre_logits=True)
# output is a (1, num_features) shaped tensor
```

## Model Comparison
| model | top1 | top5 | param_count | img_size |
| -------------------------------------------------- | ------ | ------ | ----------- | -------- |
| [vit_mediumd_patch16_reg4_gap_256.sbb_in12k_ft_in1k](https://huggingface.co/timm/vit_mediumd_patch16_reg4_gap_256.sbb_in12k_ft_in1k) | 86.202 | 97.874 | 64.11 | 256 |
| [vit_betwixt_patch16_reg4_gap_256.sbb_in12k_ft_in1k](https://huggingface.co/timm/vit_betwixt_patch16_reg4_gap_256.sbb_in12k_ft_in1k)  | 85.418 | 97.480 | 60.4 | 256 |
| [vit_medium_patch16_reg4_gap_256.sbb_in12k_ft_in1k](https://huggingface.co/timm/vit_medium_patch16_reg4_gap_256.sbb_in12k_ft_in1k)  | 84.930 | 97.386 | 38.88 | 256 |
| [vit_mediumd_patch16_rope_reg1_gap_256.sbb_in1k](https://huggingface.co/timm/vit_mediumd_patch16_rope_reg1_gap_256.sbb_in1k)  | 84.322 | 96.812 | 63.95 | 256 |            
| [vit_betwixt_patch16_rope_reg4_gap_256.sbb_in1k](https://huggingface.co/timm/vit_betwixt_patch16_rope_reg4_gap_256.sbb_in1k)  | 83.906 | 96.684 | 60.23 | 256 |
| [vit_base_patch16_rope_reg1_gap_256.sbb_in1k](https://huggingface.co/timm/vit_base_patch16_rope_reg1_gap_256.sbb_in1k)  | 83.866 | 96.67 | 86.43 | 256 |
| [vit_medium_patch16_rope_reg1_gap_256.sbb_in1k](https://huggingface.co/timm/vit_medium_patch16_rope_reg1_gap_256.sbb_in1k)  | 83.81 | 96.824 | 38.74 | 256 |
| [vit_little_patch16_reg1_gap_256.sbb_in12k_ft_in1k](https://huggingface.co/timm/vit_little_patch16_reg1_gap_256.sbb_in12k_ft_in1k)  | 83.774 | 96.972 | 22.52 | 256 |
| [vit_betwixt_patch16_reg4_gap_256.sbb_in1k](https://huggingface.co/timm/vit_betwixt_patch16_reg4_gap_256.sbb_in1k)  | 83.706 | 96.616 | 60.4 | 256 |
| [vit_betwixt_patch16_reg1_gap_256.sbb_in1k](https://huggingface.co/timm/vit_betwixt_patch16_reg1_gap_256.sbb_in1k)  | 83.628 | 96.544 | 60.4 | 256 |
| [vit_medium_patch16_reg4_gap_256.sbb_in1k](https://huggingface.co/timm/vit_medium_patch16_reg4_gap_256.sbb_in1k)  | 83.47 | 96.622 | 38.88 | 256 |
| [vit_medium_patch16_reg1_gap_256.sbb_in1k](https://huggingface.co/timm/vit_medium_patch16_reg1_gap_256.sbb_in1k)  | 83.462 | 96.548 | 38.88 | 256 |
| [vit_little_patch16_reg4_gap_256.sbb_in1k](https://huggingface.co/timm/vit_little_patch16_reg4_gap_256.sbb_in1k)  | 82.514 | 96.262 | 22.52 | 256 |
| [vit_wee_patch16_reg1_gap_256.sbb_in1k](https://huggingface.co/timm/vit_wee_patch16_reg1_gap_256.sbb_in1k)  | 80.258 | 95.360 | 13.42 | 256 |
| [vit_pwee_patch16_reg1_gap_256.sbb_in1k](https://huggingface.co/timm/vit_pwee_patch16_reg1_gap_256.sbb_in1k)  | 80.072 | 95.136 | 15.25 | 256 |

## Citation
```bibtex
@misc{rw2019timm,
  author = {Ross Wightman},
  title = {PyTorch Image Models},
  year = {2019},
  publisher = {GitHub},
  journal = {GitHub repository},
  doi = {10.5281/zenodo.4414861},
  howpublished = {\url{https://github.com/huggingface/pytorch-image-models}}
}
```
```bibtex
@article{EVA02,
  title={EVA-02: A Visual Representation for Neon Genesis},
  author={Fang, Yuxin and Sun, Quan and Wang, Xinggang and Huang, Tiejun and Wang, Xinlong and Cao, Yue},
  journal={arXiv preprint arXiv:2303.11331},
  year={2023}
}
```
```bibtex
@article{darcet2023vision,
  title={Vision Transformers Need Registers},
  author={Darcet, Timoth{'e}e and Oquab, Maxime and Mairal, Julien and Bojanowski, Piotr},
  journal={arXiv preprint arXiv:2309.16588},
  year={2023}
}
```
```bibtex
@article{dosovitskiy2020vit,
  title={An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale},
  author={Dosovitskiy, Alexey and Beyer, Lucas and Kolesnikov, Alexander and Weissenborn, Dirk and Zhai, Xiaohua and Unterthiner, Thomas and  Dehghani, Mostafa and Minderer, Matthias and Heigold, Georg and Gelly, Sylvain and Uszkoreit, Jakob and Houlsby, Neil},
  journal={ICLR},
  year={2021}
}
```