search

RIVeR-Lab / stereovoxelnet

MIT License
60 stars 7 forks source link

How to set the voxel_size and max_depth value for our own stereo dataset? #1

Closed wgh-2018 closed 1 year ago

wgh-2018 commented 1 year ago

Hello, i want to test our own stereo dataset by visualize_hie.py how to set the voxel_size and max_depth according to the camera intrinsics thank you

lhy0807 commented 1 year ago

Two lines control the max depth. In the voxel cost volume formulation https://github.com/RIVeR-Lab/stereovoxelnet/blob/08464b2e6384293d00d153deb84a6de16d3d5623/scripts/net/visualization/visualize_hie.py#L152 and in the dataset construction, like this (where the voxel size is defined) https://github.com/RIVeR-Lab/stereovoxelnet/blob/08464b2e6384293d00d153deb84a6de16d3d5623/datasets/parse_rosbag.py#L55 If you plan to test on your own stereo dataset, I recommend fine-tuning the pretrained model using your own dataset.

wgh-2018 commented 1 year ago

thanks for your reply! if i fine-tuning the pretrained model using my dataset， can i use the same max_depth and voxel_size as the DSdataset when load my dataset？my camera Intrinsics and baseline is different from DSdataset.

lhy0807 commented 1 year ago

Sure, you can use the same max_depth and voxel_size to train on your dataset. Let me know if you have more questions