Repvgg A1

🚀 RepVGG-A1 model

This model is pretrained on ImageNette. The RepVGG architecture was introduced in this paper, aiming to provide a powerful solution for image classification tasks.

🚀 Quick Start

The RepVGG-A1 model is pretrained on ImageNette and can be used for image classification. You can follow the steps below to install and use it.

✨ Features

• The core idea of the author is to distinguish the training architecture (with shortcut connections), from the inference one (a pure highway network).
• By designing the residual block, the training architecture can be reparametrized into a simple sequence of convolutions and non-linear activations.

📦 Installation

Prerequisites

Python 3.6 (or higher) and pip/conda are required to install Holocron.

Latest stable release

You can install the last stable release of the package using pypi as follows:

pip install pylocron

or using conda:

conda install -c frgfm pylocron

Developer mode

Alternatively, if you wish to use the latest features of the project that haven't made their way to a release yet, you can install the package from source (install Git first):

git clone https://github.com/frgfm/Holocron.git
pip install -e Holocron/.

💻 Usage Examples

Basic Usage

from PIL import Image
from torchvision.transforms import Compose, ConvertImageDtype, Normalize, PILToTensor, Resize
from torchvision.transforms.functional import InterpolationMode
from holocron.models import model_from_hf_hub

model = model_from_hf_hub("frgfm/repvgg_a1").eval()

img = Image.open(path_to_an_image).convert("RGB")

# Preprocessing
config = model.default_cfg
transform = Compose([
    Resize(config['input_shape'][1:], interpolation=InterpolationMode.BILINEAR),
    PILToTensor(),
    ConvertImageDtype(torch.float32),
    Normalize(config['mean'], config['std'])
])

input_tensor = transform(img).unsqueeze(0)

# Inference
with torch.inference_mode():
    output = model(input_tensor)
probs = output.squeeze(0).softmax(dim=0)

📄 License

This project is licensed under the Apache-2.0 license.

📚 Documentation

Model description

The core idea of the author is to distinguish the training architecture (with shortcut connections), from the inference one (a pure highway network). By designing the residual block, the training architecture can be reparametrized into a simple sequence of convolutions and non-linear activations.

📚 Citation

Original paper

@article{DBLP:journals/corr/abs-2101-03697,
  author    = {Xiaohan Ding and
               Xiangyu Zhang and
               Ningning Ma and
               Jungong Han and
               Guiguang Ding and
               Jian Sun},
  title     = {RepVGG: Making VGG-style ConvNets Great Again},
  journal   = {CoRR},
  volume    = {abs/2101.03697},
  year      = {2021},
  url       = {https://arxiv.org/abs/2101.03697},
  eprinttype = {arXiv},
  eprint    = {2101.03697},
  timestamp = {Tue, 09 Feb 2021 15:29:34 +0100},
  biburl    = {https://dblp.org/rec/journals/corr/abs-2101-03697.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

Source of this implementation

@software{Fernandez_Holocron_2020,
author = {Fernandez, François-Guillaume},
month = {5},
title = {{Holocron}},
url = {https://github.com/frgfm/Holocron},
year = {2020}
}