Tool for creating optical flow datasets using HoloLens 2, assuming static scenes.
This tool depends on the HoloLens positional tracking to perform depth alignment and optical flow estimation. For best results, please follow the practices described in HoloLens environment considerations.
- Install hl2ss on your HoloLens.
- Run dataset_step_01_capture_video.py to capture a video sequence comprised of RGB data from the HoloLens front camera and unaligned depth data from the HoloLens ToF sensor.
- Run dataset_step_02_generate_rgbd.py to extract RGB frames, intrinsics, extrinsics, poses, and generate aligned depth frames.
- Run dataset_step_03_generate_flow.py to generate optical flows.
- (Optional) Run dataset_step_04_generate_disparity.py to convert aligned depth frames to disparity (horizontal flow).
Repeat steps 2 through 5 for each sequence.