cspdarknet53_mish

CSP-Darknet-53 Mish model

Pretrained on ImageNette. The CSP-Darknet-53 Mish architecture was introduced in this paper.

Model description

The core idea of the author is to change the convolutional stage by adding cross stage partial blocks in the architecture and replace activations with Mish.

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 instructions

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/cspdarknet53_mish").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)

Citation

Original paper

@article{DBLP:journals/corr/abs-1911-11929,
  author    = {Chien{-}Yao Wang and
               Hong{-}Yuan Mark Liao and
               I{-}Hau Yeh and
               Yueh{-}Hua Wu and
               Ping{-}Yang Chen and
               Jun{-}Wei Hsieh},
  title     = {CSPNet: {A} New Backbone that can Enhance Learning Capability of {CNN}},
  journal   = {CoRR},
  volume    = {abs/1911.11929},
  year      = {2019},
  url       = {http://arxiv.org/abs/1911.11929},
  eprinttype = {arXiv},
  eprint    = {1911.11929},
  timestamp = {Tue, 03 Dec 2019 20:41:07 +0100},
  biburl    = {https://dblp.org/rec/journals/corr/abs-1911-11929.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}
}