Closed BigTeacher-777 closed 3 years ago
Hi,
Unfortunately, we cannot easily release our LEA data loader code. But a great starting point is to follow the LEA data loading structure for the KITTI dataset.
For the shell script, you could use the predict_sf.sh
as a template and change the flag --sceneflow=1
to --argoverse=1
(given that you've created the argoverse
option, similar to other data loaders), and set the crop flags to --crop_height=2064 --crop_width=2496
if using the full Argoverse stereo resolution.
Note that we used the full stereo resolution, and for memory reason, we run the inference on the CPU.
Can you release the data loader code of the Argverse dataset for baseline LEAStereo?