You can visualize the flows by nevigating to scripts/data
:
python viz_all.py
You may need to modify the dataset path in viz_pathes
. To visualize the simulated play dataset, use --viz_sam
. You can adjust the minimum distance a keypoint travels in the image space by specifying viz_thresholds
; only keypoints that move more than the specified threshold will be visualized.
For visualizing real-world task demonstrations, use --viz_bbox
and set viz_thresholds
to 0. To visualize the simulation dataset, use --viz_sam
and set viz_thresholds
to -1, which utilizes the existing moving mask in the dataset. You can also set viz_thresholds
to any value greater than 0 for visualization.
To display the trajectory, uncomment the --draw_line
option.
Please note that when visualizing rigid objects, you may observe some noisy flows and robot flows, as the current visualization code does not filter them out. However, these are filtered out during policy training.