Zak Murez
Zak Murez
With large images, cv2.resize() is called. https://github.com/cvg/Hierarchical-Localization/blob/99676f7a98db0b3e69dc42122de2f972f966f210/hloc/extract_features.py#L100 However this hangs the pytorch dataloader due to a threading issue (see https://stackoverflow.com/questions/54013846/pytorch-dataloader-stucked-if-using-opencv-resize-method).
Fixes typo bug in #19
fixes issue #10 and #17
I think it would be valuable to the research community to support the NuScenes dataset (https://www.nuscenes.org/nuscenes). This is a relatively large real world multi camera driving dataset popular in many...
This adds support for the NuScenes dataset #921. Optionally you can use the 3D cuboid labels to mask dynamic objects.
When using `traj=="filename"` `camera_ray_bundle` is generated on the cpu and then moved to the gpu which can be slow.
These minor changes allow the component modules to be torch scripted.