Refactoring OpenMixup to v0.2.0 according to MMSelfsup:
Update training settings and support self-supervised learning (SSL) methods (new methods, new SSL datasets, benchmarks, tools, etc.). Provide analyzing tools for self-supervised methods (e.g., visualization and linear/kNN classification metrics).
Optimize config structures to support config reusing (datasets, etc.) as MMClassification.
Fix bugs of vision transformers (reproducing results of ViT-related methods), which require mmcv-full installation.
Support self-supervised learning benchmarks on the small-scale datasets, e.g., CIFAR-10/100, ImageNet-100.
Model zoos and paper lists of awesome mixups are provided.
We will release v0.1.2 and v0.1.3 before v0.2.0. More configs and results will be provided in version v0.2.0. Configs are reorganized.
Refactoring OpenMixup to v0.2.0 according to MMSelfsup:
We will release v0.1.2 and v0.1.3 before v0.2.0. More configs and results will be provided in version v0.2.0. Configs are reorganized.