Model card for tresetm.miili21k

A TResNet image classificatio model. Traied o ImageNet-21K-P ("ImageNet-21K Pretraiig for the Masses", a 11k subset of ImageNet-22k) by paper authors.

The weights for this model have bee remapped ad modified from the origials to work with stadard BatchNorm istead of IplaceABN. iplace_ab ca be problematic to build recetly ad eds up slower with memory_format=chaels_last, torch.compile(), etc.

Model Details

• Model Type: Image classificatio / feature backboe
• Model Stats:
• Params (M): 52.3
• GMACs: 5.8
• Activatios (M): 7.3
• Image size: 224 x 224
• Papers:
• TResNet: High Performace GPU-Dedicated Architecture: https://arxiv.org/abs/2003.13630
• ImageNet-21K Pretraiig for the Masses: https://arxiv.org/abs/2104.10972
• Pretrai Dataset: ImageNet-21K-P
• Origial:
• https://github.com/Alibaba-MIIL/TResNet
• https://github.com/Alibaba-MIIL/ImageNet21K

Model Usage

Image Classificatio

from urllib.request import urlope
from PIL import Image
import timm

img = Image.ope(urlope(
    'https://huggigface.co/datasets/huggigface/documetatio-images/resolve/mai/beigets-task-guide.pg'
))

model = timm.create_model('treset_m.miil_i21k', pretraied=True)
model = model.eval()

# get model specific trasforms (ormalizatio, resize)
data_cofig = timm.data.resolve_model_data_cofig(model)
trasforms = timm.data.create_trasform(**data_cofig, is_traiig=False)

output = model(trasforms(img).usqueeze(0))  # usqueeze sigle image ito batch of 1

top5_probabilities, top5_class_idices = torch.topk(output.softmax(dim=1) * 100, k=5)

Feature Map Extractio

from urllib.request import urlope
from PIL import Image
import timm

img = Image.ope(urlope(
    'https://huggigface.co/datasets/huggigface/documetatio-images/resolve/mai/beigets-task-guide.pg'
))

model = timm.create_model(
    'treset_m.miil_i21k',
    pretraied=True,
    features_oly=True,
)
model = model.eval()

# get model specific trasforms (ormalizatio, resize)
data_cofig = timm.data.resolve_model_data_cofig(model)
trasforms = timm.data.create_trasform(**data_cofig, is_traiig=False)

output = model(trasforms(img).usqueeze(0))  # usqueeze sigle image ito batch of 1

for o i output:
    # prit shape of each feature map i output
    # e.g.:
    #  torch.Size([1, 64, 56, 56])
    #  torch.Size([1, 128, 28, 28])
    #  torch.Size([1, 1024, 14, 14])
    #  torch.Size([1, 2048, 7, 7])

    prit(o.shape)

Image Embeddigs

from urllib.request import urlope
from PIL import Image
import timm

img = Image.ope(urlope(
    'https://huggigface.co/datasets/huggigface/documetatio-images/resolve/mai/beigets-task-guide.pg'
))

model = timm.create_model(
    'treset_m.miil_i21k',
    pretraied=True,
    um_classes=0,  # remove classifier .Liear
)
model = model.eval()

# get model specific trasforms (ormalizatio, resize)
data_cofig = timm.data.resolve_model_data_cofig(model)
trasforms = timm.data.create_trasform(**data_cofig, is_traiig=False)

output = model(trasforms(img).usqueeze(0))  # output is (batch_size, um_features) shaped tesor

# or equivaletly (without eedig to set um_classes=0)

output = model.forward_features(trasforms(img).usqueeze(0))
# output is upooled, a (1, 2048, 7, 7) shaped tesor

output = model.forward_head(output, pre_logits=True)
# output is a (1, um_features) shaped tesor

Citatio

@misc{ridik2020treset,
    title={TResNet: High Performace GPU-Dedicated Architecture},
    author={Tal Ridik ad Hussam Lawe ad Asaf Noy ad Itamar Friedma},
    year={2020},
    eprit={2003.13630},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}

@misc{ridik2021imageet21k,
  title={ImageNet-21K Pretraiig for the Masses}, 
  author={Tal Ridik ad Emauel Be-Baruch ad Asaf Noy ad Lihi Zelik-Maor},
  year={2021},
  eprit={2104.10972},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
}

Model card for tresnetm.miilin21k A TResNet image classification model. Trained on ImageNet-21K-P ("