rwightman HF staff commited on
Commit
ec23d70
1 Parent(s): 77f41ef
Files changed (3) hide show
  1. README.md +109 -0
  2. config.json +36 -0
  3. pytorch_model.bin +3 -0
README.md ADDED
@@ -0,0 +1,109 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - image-classification
4
+ - timm
5
+ library_tag: timm
6
+ license: apache-2.0
7
+ datasets:
8
+ - imagenet-1k
9
+ ---
10
+ # Model card for efficientformer_l1.snap_dist_in1k
11
+
12
+ A EfficientFormer image classification model. Pretrained with distillation on ImageNet-1k.
13
+
14
+ ## Model Details
15
+ - **Model Type:** Image classification / feature backbone
16
+ - **Model Stats:**
17
+ - Params (M): 12.3
18
+ - GMACs: 1.3
19
+ - Activations (M): 5.5
20
+ - Image size: 224 x 224
21
+ - **Original:** https://github.com/snap-research/EfficientFormer
22
+ - **Papers:**
23
+ - EfficientFormer: Vision Transformers at MobileNet Speed: https://arxiv.org/abs/2206.01191
24
+ - **Dataset:** ImageNet-1k
25
+
26
+ ## Model Usage
27
+ ### Image Classification
28
+ ```python
29
+ from urllib.request import urlopen
30
+ from PIL import Image
31
+ import timm
32
+
33
+ img = Image.open(
34
+ urlopen('https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'))
35
+
36
+ model = timm.create_model('efficientformer_l1.snap_dist_in1k', pretrained=True)
37
+ model = model.eval()
38
+
39
+ # get model specific transforms (normalization, resize)
40
+ data_config = timm.data.resolve_model_data_config(model)
41
+ transforms = timm.data.create_transform(**data_config, is_training=False)
42
+
43
+ output = model(transforms(img).unsqueeze(0)) # unsqueeze single image into batch of 1
44
+
45
+ top5_probabilities, top5_class_indices = torch.topk(output.softmax(dim=1) * 100, k=5)
46
+ ```
47
+
48
+ ### Image Embeddings
49
+ ```python
50
+ from urllib.request import urlopen
51
+ from PIL import Image
52
+ import timm
53
+
54
+ img = Image.open(
55
+ urlopen('https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'))
56
+
57
+ model = timm.create_model(
58
+ 'efficientformer_l1.snap_dist_in1k',
59
+ pretrained=True,
60
+ num_classes=0, # remove classifier nn.Linear
61
+ )
62
+ model = model.eval()
63
+
64
+ # get model specific transforms (normalization, resize)
65
+ data_config = timm.data.resolve_model_data_config(model)
66
+ transforms = timm.data.create_transform(**data_config, is_training=False)
67
+
68
+ output = model(transforms(img).unsqueeze(0)) # output is (batch_size, num_features) shaped tensor
69
+
70
+ # or equivalently (without needing to set num_classes=0)
71
+
72
+ output = model.forward_features(transforms(img).unsqueeze(0))
73
+ # output is unpooled (ie.e a (batch_size, num_features, H, W) tensor
74
+
75
+ output = model.forward_head(output, pre_logits=True)
76
+ # output is (batch_size, num_features) tensor
77
+ ```
78
+
79
+ ## Model Comparison
80
+ |model |top1 |top5 |param_count|img_size|
81
+ |-----------------------------------|------|------|-----------|--------|
82
+ |efficientformerv2_l.snap_dist_in1k |83.628|96.54 |26.32 |224 |
83
+ |efficientformer_l7.snap_dist_in1k |83.368|96.534|82.23 |224 |
84
+ |efficientformer_l3.snap_dist_in1k |82.572|96.24 |31.41 |224 |
85
+ |efficientformerv2_s2.snap_dist_in1k|82.128|95.902|12.71 |224 |
86
+ |efficientformer_l1.snap_dist_in1k |80.496|94.984|12.29 |224 |
87
+ |efficientformerv2_s1.snap_dist_in1k|79.698|94.698|6.19 |224 |
88
+ |efficientformerv2_s0.snap_dist_in1k|76.026|92.77 |3.6 |224 |
89
+
90
+ ## Citation
91
+ ```bibtex
92
+ @article{li2022efficientformer,
93
+ title={EfficientFormer: Vision Transformers at MobileNet Speed},
94
+ author={Li, Yanyu and Yuan, Geng and Wen, Yang and Hu, Ju and Evangelidis, Georgios and Tulyakov, Sergey and Wang, Yanzhi and Ren, Jian},
95
+ journal={arXiv preprint arXiv:2206.01191},
96
+ year={2022}
97
+ }
98
+ ```
99
+ ```bibtex
100
+ @misc{rw2019timm,
101
+ author = {Ross Wightman},
102
+ title = {PyTorch Image Models},
103
+ year = {2019},
104
+ publisher = {GitHub},
105
+ journal = {GitHub repository},
106
+ doi = {10.5281/zenodo.4414861},
107
+ howpublished = {\url{https://github.com/rwightman/pytorch-image-models}}
108
+ }
109
+ ```
config.json ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architecture": "efficientformer_l1",
3
+ "num_classes": 1000,
4
+ "num_features": 448,
5
+ "global_pool": "avg",
6
+ "pretrained_cfg": {
7
+ "tag": "snap_dist_in1k",
8
+ "custom_load": false,
9
+ "input_size": [
10
+ 3,
11
+ 224,
12
+ 224
13
+ ],
14
+ "fixed_input_size": true,
15
+ "interpolation": "bicubic",
16
+ "crop_pct": 0.95,
17
+ "crop_mode": "center",
18
+ "mean": [
19
+ 0.485,
20
+ 0.456,
21
+ 0.406
22
+ ],
23
+ "std": [
24
+ 0.229,
25
+ 0.224,
26
+ 0.225
27
+ ],
28
+ "num_classes": 1000,
29
+ "pool_size": null,
30
+ "first_conv": "stem.conv1",
31
+ "classifier": [
32
+ "head",
33
+ "head_dist"
34
+ ]
35
+ }
36
+ }
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:33fac87c4f7e0255d08b4ce9471ab10a87bb5651695e1efdaf0b00b1ca923a68
3
+ size 49405489