Model card for haloregnetz_b.ra3_in1k

A HaloNet image classification model (, based on RegNet-Z architecture). Trained on ImageNet-1k in timm by Ross Wightman.

NOTE: this model did not adhere to any specific paper configuration, it was tuned for reasonable training times and reduced frequency of self-attention blocks.

Recipe details:

• RandAugment RA3 recipe. Inspired by and evolved from EfficientNet RandAugment recipes. Published as B recipe in ResNet Strikes Back.
• RMSProp (TF 1.0 behaviour) optimizer, EMA weight averaging
• Step (exponential decay w/ staircase) LR schedule with warmup

This model architecture is implemented using timm's flexible BYOBNet (Bring-Your-Own-Blocks Network).

BYOB (with BYOANet attention specific blocks) allows configuration of:

• block / stage layout
• block-type interleaving
• stem layout
• output stride (dilation)
• activation and norm layers
• channel and spatial / self-attention layers

...and also includes timm features common to many other architectures, including:

• stochastic depth
• gradient checkpointing
• layer-wise LR decay
• per-stage feature extraction

Model Details

• Model Type: Image classification / feature backbone
• Model Stats:
  • Params (M): 11.7
  • GMACs: 2.0
  • Activations (M): 11.9
  • Image size: 224 x 224
• Papers:
  • Scaling Local Self-Attention for Parameter Efficient Visual Backbones: https://arxiv.org/abs/2103.12731
  • ResNet strikes back: An improved training procedure in timm: https://arxiv.org/abs/2110.00476
• Dataset: ImageNet-1k

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('haloregnetz_b.ra3_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(
    'haloregnetz_b.ra3_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, 32, 112, 112])
    #  torch.Size([1, 48, 56, 56])
    #  torch.Size([1, 96, 28, 28])
    #  torch.Size([1, 192, 14, 14])
    #  torch.Size([1, 1536, 7, 7])

    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(
    'haloregnetz_b.ra3_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, 1536, 7, 7) 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

@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}}
}

@article{Vaswani2021ScalingLS,
  title={Scaling Local Self-Attention for Parameter Efficient Visual Backbones},
  author={Ashish Vaswani and Prajit Ramachandran and A. Srinivas and Niki Parmar and Blake A. Hechtman and Jonathon Shlens},
  journal={2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2021},
  pages={12889-12899}
}

@inproceedings{wightman2021resnet,
  title={ResNet strikes back: An improved training procedure in timm},
  author={Wightman, Ross and Touvron, Hugo and Jegou, Herve},
  booktitle={NeurIPS 2021 Workshop on ImageNet: Past, Present, and Future}
}