search

megvii-research / RealFlow

The official implementation of the ECCV 2022 Oral paper: RealFlow: EM-based Realistic Optical Flow Dataset Generation from Videos
Other
93 stars 11 forks source link

Which data augmentation is applied to train RAFT on RF-Ktrain? #4

Closed Byungki-Kwon closed 1 year ago

Byungki-Kwon commented 1 year ago

Thanks for releasing the data generation code of excellent work.

I wonder the range of data augmentation to train RAFT on RF-Ktrain, such that 1) crop_size 2) min_scale 3) max_scale 4) do_flip, which are included RAFT/core/dataset.py file.

coolbeam commented 1 year ago

Thank you. We use the official code of RAFT without any modification to train RAFT on our generated datasets. Therefore, the range of data augmentation for training RAFT on RF-Ktrain is the same as that included in RAFT/core/dataset.py file.