sscls is an image classification codebase for research related to self-supervised representation learning, written in PyTorch. sscls is based on pycls.
The goal of sscls is to provide a high-quality, high-performance codebase for self-supervised research. It is designed to be simple and flexible in order to support rapid implementation and evaluation of research ideas.
The codebase implements efficient single-machine multi-gpu training, powered by PyTorch distributed package. sscls includes implementations of standard baseline models (ResNet, ResNeXt, EfficientNet) and generic modeling functionality that can be useful for experimenting with network design. Additional models can be easily implemented. Apart from basic classification models, it has also included state-of-the-art MoCo v1&v2.
Please see INSTALL.md
for installation instructions.
After installation, please see GETTING_STARTED.md
for basic instructions on training and evaluation with pycls.
- resnet-50 epoch: 100, top1_err: 23.484001, top5_err: 6.814002
sscls is released under the MIT license. See the LICENSE file for more information.