PyTorch implementation of ST-DiODE for spatiotemporal sequence prediction.
This repository contains a cleaned and training-ready version of the project code, including:
- ST-DiODE backbone
- training / validation / testing pipeline
- custom
.npysequence dataloader - logging, checkpoint saving, and metric evaluation
The current pipeline is directly adapted to the polished codebase and is especially convenient for custom sequence data in .npy format with shape:
[N, T, C, H, W]