WeiChow/moge_l_vit

same architecture with timm/vit_large_patch14_dinov2.lvd142m

git clone https://github.com/microsoft/MoGe.git
cd MoGe

translate

import torch
from moge.model import MoGeModel

device = torch.device("cuda")
model = MoGeModel.from_pretrained("Ruicheng/moge-vitl").to(device)

# -------------------
backbone_state_dict = model.backbone.state_dict()
filtered_state_dict = {k: v for k, v in backbone_state_dict.items() if 'mask_token' not in k}
torch.save(filtered_state_dict, "pytorch_model.bin")
# -------------------

usage

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(
    'hf_hub:WeiChow/moge_l_vit',
    pretrained=True,
    num_classes=0,  # remove classifier nn.Linear
)

# for name, param in model.named_parameters():
#     print(f"Parameter: {name} - Size: {param.size()} - Total Elements: {param.numel()}")
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, 1374, 1024) shaped tensor

output = model.forward_head(output, pre_logits=True)
print(output)

Copyright saved.