This release includes several significant new features, bug fixes and tutorials
New additions include the following,
Models
- Multi-Scale Vision Transformers (MViT) along with it's model builders and associated pre-trained models in the model zoo.
MViT is a new state of the art vision transformer model that beats the existing baselines all while requiring lesser compute resources. - Audio Visual SlowFast Model - This would enable you to work with Audio and Video Modalities simultaneously
- Video action detection Resnet model and associated pre-trained models in the model zoo.
Transform
Simply, adding AugMix and MixUp transforms to your existing training recipes should boost your models baseline accuracy.
Datasets
- Ava dataset and it's associated benchmarks and tutorials