eva02_base_patch14_224.mim_in22k

An EVA02 feature / representation model. Pretrained on ImageNet-22k with masked image modeling (using EVA-CLIP as a MIM teacher) by paper authors.

EVA-02 models are vision transformers with mean pooling, SwiGLU, Rotary Position Embeddings (ROPE), and extra LN in MLP (for Base & Large).

NOTE: timm checkpoints are float32 for consistency with other models. Original checkpoints are float16 or bfloat16 in some cases, see originals if that's preferred.

Model Details

• Model Type: Image classification / feature backbone
• Model Stats:
  • Params (M): 85.8
  • GMACs: 23.2
  • Activations (M): 36.6
  • Image size: 224 x 224
• Papers:
  • EVA-02: A Visual Representation for Neon Genesis: https://arxiv.org/abs/2303.11331
  • EVA-CLIP: Improved Training Techniques for CLIP at Scale: https://arxiv.org/abs/2303.15389
• Original:
  • https://github.com/baaivision/EVA
  • https://huggingface.co/Yuxin-CV/EVA-02
• Pretrain Dataset: ImageNet-22k

Model Usage

Image Classification

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('eva02_base_patch14_224.mim_in22k', 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)

Image Embeddings

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(
    'eva02_base_patch14_224.mim_in22k',
    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, 768) shaped tensor

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

Model Comparison

Explore the dataset and runtime metrics of this model in timm model results.

Citation

@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}
}

@article{EVA-CLIP,
  title={EVA-02: A Visual Representation for Neon Genesis},
  author={Sun, Quan and Fang, Yuxin and Wu, Ledell and Wang, Xinlong and Cao, Yue},
  journal={arXiv preprint arXiv:2303.15389},
  year={2023}
}

@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}}
}