Support for KITTI-Format Datasets (RGB + Lidar) in DPVO

[akramsalim]

I am trying to use dpco with my Zeki dataset, which follows the KITTI format (RGB images + Lidar point clouds in .bin). However, I see that dpvo was trained on TartanAir, which provides RGB + dense depth maps, leading to a format mismatch.

Questions:

1. Does dpvo support KITTI-style datasets (RGB + sparse Lidar)?

2. If not, what’s the best way to convert sparse Lidar into dense depth maps? Or maybe I should consider doing another thing?

3. Are there specific depth format requirements for compatibility?

What I Tried:

Projected Lidar to depth maps but they are sparse.

Considered interpolation but unsure if this is the right approach.

Would appreciate any guidance on preparing KITTI-style data for DPVO. Thanks!