Views
No views yet
1from urllib.request import urlopen
2from PIL import Image
3import timm
4
5img = Image.open(urlopen(
6 'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
7))
8
9model = timm.create_model('convnext_small.eupe_lvd1689m', pretrained=True)
10model = model.eval()
11
12# get model specific transforms (normalization, resize)
13data_config = timm.data.resolve_model_data_config(model)
14transforms = timm.data.create_transform(**data_config, is_training=False)
15
16output = model(transforms(img).unsqueeze(0)) # unsqueeze single image into batch of 1
17
18top5_probabilities, top5_class_indices = torch.topk(output.softmax(dim=1) * 100, k=5)1from urllib.request import urlopen
2from PIL import Image
3import timm
4
5img = Image.open(urlopen(
6 'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
7))
8
9model = timm.create_model(
10 'convnext_small.eupe_lvd1689m',
11 pretrained=True,
12 features_only=True,
13)
14model = model.eval()
15
16# get model specific transforms (normalization, resize)
17data_config = timm.data.resolve_model_data_config(model)
18transforms = timm.data.create_transform(**data_config, is_training=False)
19
20output = model(transforms(img).unsqueeze(0)) # unsqueeze single image into batch of 1
21
22for o in output:
23 # print shape of each feature map in output
24 # e.g.:
25 # torch.Size([1, 96, 64, 64])
26 # torch.Size([1, 192, 32, 32])
27 # torch.Size([1, 384, 16, 16])
28 # torch.Size([1, 768, 8, 8])
29
30 print(o.shape)1from urllib.request import urlopen
2from PIL import Image
3import timm
4
5img = Image.open(urlopen(
6 'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
7))
8
9model = timm.create_model(
10 'convnext_small.eupe_lvd1689m',
11 pretrained=True,
12 num_classes=0, # remove classifier nn.Linear
13)
14model = model.eval()
15
16# get model specific transforms (normalization, resize)
17data_config = timm.data.resolve_model_data_config(model)
18transforms = timm.data.create_transform(**data_config, is_training=False)
19
20output = model(transforms(img).unsqueeze(0)) # output is (batch_size, num_features) shaped tensor
21
22# or equivalently (without needing to set num_classes=0)
23
24output = model.forward_features(transforms(img).unsqueeze(0))
25# output is unpooled, a (1, 768, 8, 8) shaped tensor
26
27output = model.forward_head(output, pre_logits=True)
28# output is a (1, num_features) shaped tensor| Model | Params | TextVQA | SQA | Realworld | POPE | GQA | MMEp | SPair | NYUv2 | ADE20k |
|---|---|---|---|---|---|---|---|---|---|---|
| EUPE-ConvNeXt-T | 29M | 43.7 | 68.8 | 47.9 | 83.4 | 63.0 | 1278.1 | 41.3 | 0.430 | 43.5 |
| EUPE-ConvNeXt-S | 50M | 45.0 | 68.9 | 50.5 | 84.0 | 64.7 | 1284.2 | 40.1 | 0.388 | 46.8 |
| EUPE-ConvNeXt-B | 89M | 46.4 | 70.1 | 53.3 | 84.7 | 65.8 | 1348.9 | 37.7 | 0.365 | 48.9 |
1@misc{zhu2026eupe,
2 title={Efficient Universal Perception Encoder},
3 author={Zhu, Chenchen and Suri, Saksham and Jose, Cijo and Oquab, Maxime and Szafraniec, Marc and Wen, Wei and Xiong, Yunyang and Labatut, Patrick and Bojanowski, Piotr and Krishnamoorthi, Raghuraman and Chandra, Vikas},
4 year={2026},
5 eprint={2603.22387},
6 archivePrefix={arXiv},
7 primaryClass={cs.CV},
8 url={https://arxiv.org/abs/2603.22387},
9}1@article{liu2022convnet,
2 author = {Zhuang Liu and Hanzi Mao and Chao-Yuan Wu and Christoph Feichtenhofer and Trevor Darrell and Saining Xie},
3 title = {A ConvNet for the 2020s},
4 journal = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
5 year = {2022},
6}1@misc{rw2019timm,
2 author = {Ross Wightman},
3 title = {PyTorch Image Models},
4 year = {2019},
5 publisher = {GitHub},
6 journal = {GitHub repository},
7 doi = {10.5281/zenodo.4414861},
8 howpublished = {\url{https://github.com/huggingface/pytorch-image-models}}
9}