- Advanced spatial augmentations borrowed from [VCN ](https://github.com/gengshan-y/VCN/blob/master/dataloader/flow_transforms.py) and [AutoFlow](https://github.com/google-research/opticalflow-autoflow/blob/main/src/dataset_lib/augmentations/aug_params.py)
- Added Normalization as a configurable dataset parameter
- PWC-Net and FlowNetC bug fix:MultiScale Loss and Correlation Computation
- PWC-Net: decouple PWC-Net modules
- Distributed Training `barrier` synchronization
- Kubric dataset support [Kubric optical flow](https://github.com/google-research/kubric/tree/main/challenges/optical_flow)
An additional script is required to convert `.tfrecords` to PyTorch-compatible DataLoader. This script is provided in a standalone repository [kubric-flow](https://github.com/prajnan93/kubric-flow) as it requires installation of tensorflow.