timm
/

Image Classification
timm
PyTorch
Safetensors
File size: 16,073 Bytes
ddc0372
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b4e87f0
ddc0372
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
---
tags:
- image-classification
- timm
library_name: timm
license: apache-2.0
datasets:
- imagenet-1k
---
# Model card for resnetv2_34d.ra4_e3600_r224_in1k

A ResNet image classification model. Trained on ImageNet-1k by Ross Wightman.

Trained with `timm` scripts using hyper-parameters inspired by the MobileNet-V4 small, mixed with go-to hparams from `timm` and "ResNet Strikes Back".

A collection of hparams (timm .yaml config files) for this training series can be found here: https://gist.github.com/rwightman/f6705cb65c03daeebca8aa129b1b94ad

## Model Details
- **Model Type:** Image classification / feature backbone
- **Model Stats:**
  - Params (M): 21.8
  - GMACs: 3.9
  - Activations (M): 4.5
  - Image size: train = 224 x 224, test = 288 x 288
- **Dataset:** ImageNet-1k
- **Papers:**
  - PyTorch Image Models: https://github.com/huggingface/pytorch-image-models
  - ResNet strikes back: An improved training procedure in timm: https://arxiv.org/abs/2110.00476
  - Deep Residual Learning for Image Recognition: https://arxiv.org/abs/1512.03385
  - MobileNetV4 -- Universal Models for the Mobile Ecosystem: https://arxiv.org/abs/2404.10518

## 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('resnetv2_34d.ra4_e3600_r224_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(
    'resnetv2_34d.ra4_e3600_r224_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, 64, 112, 112])
    #  torch.Size([1, 64, 56, 56])
    #  torch.Size([1, 128, 28, 28])
    #  torch.Size([1, 256, 14, 14])
    #  torch.Size([1, 512, 7, 7])

    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(
    'resnetv2_34d.ra4_e3600_r224_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, 512, 7, 7) shaped tensor

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

## Model Comparison
### By Top-1

| model                                                                                                                    | top1   | top5   | param_count | img_size |
|--------------------------------------------------------------------------------------------------------------------------|--------|--------|-------------|----------|
| [mobilenetv4_conv_aa_large.e230_r448_in12k_ft_in1k](http://hf.co/timm/mobilenetv4_conv_aa_large.e230_r448_in12k_ft_in1k) | 84.99  | 97.294 | 32.59       | 544      |
| [mobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k](http://hf.co/timm/mobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k) | 84.772 | 97.344 | 32.59       | 480      |
| [mobilenetv4_conv_aa_large.e230_r448_in12k_ft_in1k](http://hf.co/timm/mobilenetv4_conv_aa_large.e230_r448_in12k_ft_in1k) | 84.64  | 97.114 | 32.59       | 448      |
| [mobilenetv4_hybrid_large.ix_e600_r384_in1k](http://hf.co/timm/mobilenetv4_hybrid_large.ix_e600_r384_in1k)               | 84.356 | 96.892 | 37.76       | 448      |
| [mobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k](http://hf.co/timm/mobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k) | 84.314 | 97.102 | 32.59       | 384      |
| [mobilenetv4_hybrid_large.e600_r384_in1k](http://hf.co/timm/mobilenetv4_hybrid_large.e600_r384_in1k)                     | 84.266 | 96.936 | 37.76       | 448      |
| [mobilenetv4_hybrid_large.ix_e600_r384_in1k](http://hf.co/timm/mobilenetv4_hybrid_large.ix_e600_r384_in1k)               | 83.990 | 96.702 | 37.76       | 384      |
| [mobilenetv4_conv_aa_large.e600_r384_in1k](http://hf.co/timm/mobilenetv4_conv_aa_large.e600_r384_in1k)                   | 83.824 | 96.734 | 32.59       | 480      |
| [mobilenetv4_hybrid_large.e600_r384_in1k](http://hf.co/timm/mobilenetv4_hybrid_large.e600_r384_in1k)                     | 83.800 | 96.770 | 37.76       | 384      |
| [mobilenetv4_hybrid_medium.ix_e550_r384_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.ix_e550_r384_in1k)             | 83.394 | 96.760 | 11.07       | 448      |
| [mobilenetv4_conv_large.e600_r384_in1k](http://hf.co/timm/mobilenetv4_conv_large.e600_r384_in1k)                         | 83.392 | 96.622 | 32.59       | 448      |
| [mobilenetv4_conv_aa_large.e600_r384_in1k](http://hf.co/timm/mobilenetv4_conv_aa_large.e600_r384_in1k)                   | 83.244 | 96.392 | 32.59       | 384      |
| [mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k) | 82.99  | 96.67  | 11.07       | 320      |
| [mobilenetv4_hybrid_medium.ix_e550_r384_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.ix_e550_r384_in1k)             | 82.968 | 96.474 | 11.07       | 384      |
| [mobilenetv4_conv_large.e600_r384_in1k](http://hf.co/timm/mobilenetv4_conv_large.e600_r384_in1k)                         | 82.952 | 96.266 | 32.59       | 384      |
| [mobilenetv4_conv_large.e500_r256_in1k](http://hf.co/timm/mobilenetv4_conv_large.e500_r256_in1k)                         | 82.674 | 96.31  | 32.59       | 320      |
| [mobilenetv4_hybrid_medium.ix_e550_r256_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.ix_e550_r256_in1k)             | 82.492 | 96.278 | 11.07       | 320      |
| [mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k) | 82.364 | 96.256 | 11.07       | 256      |
| [mobilenetv4_conv_large.e500_r256_in1k](http://hf.co/timm/mobilenetv4_conv_large.e500_r256_in1k)                         | 81.862 | 95.69  | 32.59       | 256      |
| [resnet50d.ra4_e3600_r224_in1k](http://hf.co/timm/resnet50d.ra4_e3600_r224_in1k)                                         | 81.838 | 95.922 | 25.58       | 288      |    
| [mobilenetv3_large_150d.ra4_e3600_r256_in1k](http://hf.co/timm/mobilenetv3_large_150d.ra4_e3600_r256_in1k)               | 81.806 | 95.9   | 14.62       | 320      |
| [mobilenetv4_hybrid_medium.ix_e550_r256_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.ix_e550_r256_in1k)             | 81.446 | 95.704 | 11.07       | 256      |
| [efficientnet_b1.ra4_e3600_r240_in1k](http://hf.co/timm/efficientnet_b1.ra4_e3600_r240_in1k)                             | 81.440 | 95.700 | 7.79        | 288      |
| [mobilenetv4_hybrid_medium.e500_r224_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.e500_r224_in1k)                   | 81.276 | 95.742 | 11.07       | 256      |
| [resnet50d.ra4_e3600_r224_in1k](http://hf.co/timm/resnet50d.ra4_e3600_r224_in1k)                                         | 80.952 | 95.384 | 25.58       | 224      |
| [mobilenetv3_large_150d.ra4_e3600_r256_in1k](http://hf.co/timm/mobilenetv3_large_150d.ra4_e3600_r256_in1k)               | 80.944 | 95.448 | 14.62       | 256      |
| [mobilenetv4_conv_medium.e500_r256_in1k](http://hf.co/timm/mobilenetv4_conv_medium.e500_r256_in1k)                       | 80.858 | 95.768 | 9.72        | 320      |
| [mobilenet_edgetpu_v2_m.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenet_edgetpu_v2_m.ra4_e3600_r224_in1k)               | 80.680 | 95.442 | 8.46        | 256      |
| [mobilenetv4_hybrid_medium.e500_r224_in1k](http://hf.co/timm/mobilenetv4_hybrid_medium.e500_r224_in1k)                   | 80.442 | 95.38  | 11.07       | 224      |
| [efficientnet_b1.ra4_e3600_r240_in1k](http://hf.co/timm/efficientnet_b1.ra4_e3600_r240_in1k)                             | 80.406 | 95.152 | 7.79        | 240      |
| [mobilenetv4_conv_blur_medium.e500_r224_in1k](http://hf.co/timm/mobilenetv4_conv_blur_medium.e500_r224_in1k)             | 80.142 | 95.298 | 9.72        | 256      |
| [mobilenet_edgetpu_v2_m.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenet_edgetpu_v2_m.ra4_e3600_r224_in1k)               | 80.130 | 95.002 | 8.46        | 224      |
| [mobilenetv4_conv_medium.e500_r256_in1k](http://hf.co/timm/mobilenetv4_conv_medium.e500_r256_in1k)                       | 79.928 | 95.184 | 9.72        | 256      |
| [mobilenetv4_conv_medium.e500_r224_in1k](http://hf.co/timm/mobilenetv4_conv_medium.e500_r224_in1k)                       | 79.808 | 95.186 | 9.72        | 256      |
| [resnetv2_34d.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_34d.ra4_e3600_r224_in1k)                                   | 79.590 | 94.770 | 21.82       | 288      |
| [mobilenetv4_conv_blur_medium.e500_r224_in1k](http://hf.co/timm/mobilenetv4_conv_blur_medium.e500_r224_in1k)             | 79.438 | 94.932 | 9.72        | 224      |
| [efficientnet_b0.ra4_e3600_r224_in1k](http://hf.co/timm/efficientnet_b0.ra4_e3600_r224_in1k)                             | 79.364 | 94.754 | 5.29        | 256      |
| [mobilenetv4_conv_medium.e500_r224_in1k](http://hf.co/timm/mobilenetv4_conv_medium.e500_r224_in1k)                       | 79.094 | 94.77  | 9.72        | 224      |
| [resnetv2_34.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_34.ra4_e3600_r224_in1k)                                     | 79.072 | 94.566 | 21.80       | 288      |
| [resnet34.ra4_e3600_r224_in1k](http://hf.co/timm/resnet34.ra4_e3600_r224_in1k)                                           | 78.952 | 94.450 | 21.80       | 288      |
| [efficientnet_b0.ra4_e3600_r224_in1k](http://hf.co/timm/efficientnet_b0.ra4_e3600_r224_in1k)                             | 78.584 | 94.338 | 5.29        | 224      |
| [resnetv2_34d.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_34d.ra4_e3600_r224_in1k)                                   | 78.268 | 93.952 | 21.82       | 224      |
| [resnetv2_34.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_34.ra4_e3600_r224_in1k)                                     | 77.636 | 93.528 | 21.80       | 224      |
| [mobilenetv1_125.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv1_125.ra4_e3600_r224_in1k)                             | 77.600 | 93.804 | 6.27        | 256      |
| [resnet34.ra4_e3600_r224_in1k](http://hf.co/timm/resnet34.ra4_e3600_r224_in1k)                                           | 77.448 | 93.502 | 21.80       | 224      |
| [mobilenetv3_large_100.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv3_large_100.ra4_e3600_r224_in1k)                 | 77.164 | 93.336 | 5.48        | 256      |              
| [mobilenetv1_125.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv1_125.ra4_e3600_r224_in1k)                             | 76.924 | 93.234 | 6.27        | 224      |
| [mobilenetv1_100h.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv1_100h.ra4_e3600_r224_in1k)                           | 76.596 | 93.272 | 5.28        | 256      |
| [mobilenetv3_large_100.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv3_large_100.ra4_e3600_r224_in1k)                 | 76.310 | 92.846 | 5.48        | 224      |
| [mobilenetv1_100.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv1_100.ra4_e3600_r224_in1k)                             | 76.094 | 93.004 | 4.23        | 256      |
| [resnetv2_18d.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_18d.ra4_e3600_r224_in1k)                                   | 76.044 | 93.020 | 11.71       | 288      |
| [resnet18d.ra4_e3600_r224_in1k](http://hf.co/timm/resnet18d.ra4_e3600_r224_in1k)                                         | 76.024 | 92.780 | 11.71       | 288      |
| [mobilenetv1_100h.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv1_100h.ra4_e3600_r224_in1k)                           | 75.662 | 92.504 | 5.28        | 224      |
| [mobilenetv1_100.ra4_e3600_r224_in1k](http://hf.co/timm/mobilenetv1_100.ra4_e3600_r224_in1k)                             | 75.382 | 92.312 | 4.23        | 224      |
| [resnetv2_18.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_18.ra4_e3600_r224_in1k)                                     | 75.340 | 92.678 | 11.69       | 288      |
| [mobilenetv4_conv_small.e2400_r224_in1k](http://hf.co/timm/mobilenetv4_conv_small.e2400_r224_in1k)                       | 74.616 | 92.072 | 3.77        | 256      |      
| [resnetv2_18d.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_18d.ra4_e3600_r224_in1k)                                   | 74.412 | 91.936 | 11.71       | 224      |
| [resnet18d.ra4_e3600_r224_in1k](http://hf.co/timm/resnet18d.ra4_e3600_r224_in1k)                                         | 74.322 | 91.832 | 11.71       | 224      |
| [mobilenetv4_conv_small.e1200_r224_in1k](http://hf.co/timm/mobilenetv4_conv_small.e1200_r224_in1k)                       | 74.292 | 92.116 | 3.77        | 256      |
| [mobilenetv4_conv_small.e2400_r224_in1k](http://hf.co/timm/mobilenetv4_conv_small.e2400_r224_in1k)                       | 73.756 | 91.422 | 3.77        | 224      |
| [resnetv2_18.ra4_e3600_r224_in1k](http://hf.co/timm/resnetv2_18.ra4_e3600_r224_in1k)                                     | 73.578 | 91.352 | 11.69       | 224      |
| [mobilenetv4_conv_small.e1200_r224_in1k](http://hf.co/timm/mobilenetv4_conv_small.e1200_r224_in1k)                       | 73.454 | 91.34  | 3.77        | 224      |
| [mobilenetv4_conv_small_050.e3000_r224_in1k](http://hf.co/timm/mobilenetv4_conv_small_050.e3000_r224_in1k)               | 65.810 | 86.424 | 2.24        | 256      |
| [mobilenetv4_conv_small_050.e3000_r224_in1k](http://hf.co/timm/mobilenetv4_conv_small_050.e3000_r224_in1k)               | 64.762 | 85.514 | 2.24        | 224      |

## 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
@inproceedings{wightman2021resnet,
  title={ResNet strikes back: An improved training procedure in timm},
  author={Wightman, Ross and Touvron, Hugo and Jegou, Herve},
  booktitle={NeurIPS 2021 Workshop on ImageNet: Past, Present, and Future}
}
```
```bibtex
@article{He2015,
  author = {Kaiming He and Xiangyu Zhang and Shaoqing Ren and Jian Sun},
  title = {Deep Residual Learning for Image Recognition},
  journal = {arXiv preprint arXiv:1512.03385},
  year = {2015}
}
```
```bibtex
@article{qin2024mobilenetv4,
  title={MobileNetV4-Universal Models for the Mobile Ecosystem},
  author={Qin, Danfeng and Leichner, Chas and Delakis, Manolis and Fornoni, Marco and Luo, Shixin and Yang, Fan and Wang, Weijun and Banbury, Colby and Ye, Chengxi and Akin, Berkin and others},
  journal={arXiv preprint arXiv:2404.10518},
  year={2024}
}
```