Model card for vit_base_patch16_rope_224.naver_in1k

A ROPE-ViT model with ROPE for image classification. Trained on ImageNet-1k by paper authors.

Model Details

• Model Type: Image classification / feature backbone
• Model Stats:
  • Params (M): 86.4
  • GMACs: 17.6
  • Activations (M): 23.9
  • Image size: 224 x 224
• Dataset: ImageNet-1k
• Papers:
  • Rotary Position Embedding for Vision Transformer: https://arxiv.org/abs/2403.13298
• Original: https://github.com/naver-ai/rope-vit

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('vit_base_patch16_rope_224.naver_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

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_base_patch16_rope_224.naver_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, 768, 14, 14])
    #  torch.Size([1, 768, 14, 14])
    #  torch.Size([1, 768, 14, 14])

    print(o.shape)

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(
    'vit_base_patch16_rope_224.naver_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, 197, 768) shaped tensor

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

Model Comparison

By Top-1 (224x224 resolution)

model	img_size	top1	top5	param_count
vit_large_patch16_rope_mixed_ape_224.naver_in1k	224	84.84	97.122	304.4
vit_large_patch16_rope_mixed_224.naver_in1k	224	84.828	97.116	304.2
vit_large_patch16_rope_ape_224.naver_in1k	224	84.65	97.154	304.37
vit_large_patch16_rope_224.naver_in1k	224	84.648	97.122	304.17
vit_base_patch16_rope_mixed_ape_224.naver_in1k	224	83.894	96.754	86.59
vit_base_patch16_rope_mixed_224.naver_in1k	224	83.804	96.712	86.44
vit_base_patch16_rope_ape_224.naver_in1k	224	83.782	96.61	86.59
vit_base_patch16_rope_224.naver_in1k	224	83.718	96.672	86.43
vit_small_patch16_rope_224.naver_in1k	224	81.23	95.022	21.98
vit_small_patch16_rope_mixed_224.naver_in1k	224	81.216	95.022	21.99
vit_small_patch16_rope_ape_224.naver_in1k	224	81.004	95.016	22.06
vit_small_patch16_rope_mixed_ape_224.naver_in1k	224	80.986	94.976	22.06

Extrapolation Performance (320x320 resolution)

model	img_size	top1	top5	param_count
vit_large_patch16_rope_mixed_224.naver_in1k	320	85.656	97.474	304.2
vit_large_patch16_rope_mixed_ape_224.naver_in1k	320	85.594	97.508	304.4
vit_large_patch16_rope_ape_224.naver_in1k	320	85.344	97.438	304.37
vit_large_patch16_rope_224.naver_in1k	320	85.258	97.42	304.17
vit_base_patch16_rope_mixed_224.naver_in1k	320	84.65	97.106	86.44
vit_base_patch16_rope_mixed_ape_224.naver_in1k	320	84.58	97.144	86.59
vit_base_patch16_rope_ape_224.naver_in1k	320	84.368	96.968	86.59
vit_base_patch16_rope_224.naver_in1k	320	84.296	96.898	86.43
vit_small_patch16_rope_mixed_224.naver_in1k	320	82.238	95.592	21.99
vit_small_patch16_rope_mixed_ape_224.naver_in1k	320	82.056	95.586	22.06
vit_small_patch16_rope_ape_224.naver_in1k	320	81.944	95.506	22.06
vit_small_patch16_rope_224.naver_in1k	320	81.46	95.142	21.98

ROPE-ViT demonstrates strong extrapolation performance when tested at higher resolutions than training resolution (224x224), maintaining or improving accuracy at 320x320.

Citation

@inproceedings{heo2024rotary,
  title={Rotary position embedding for vision transformer},
  author={Heo, Byeongho and Park, Song and Han, Dongyoon and Yun, Sangdoo},
  booktitle={European Conference on Computer Vision},
  pages={289--305},
  year={2024},
  organization={Springer}
}