search

Sungmin-Woo / ProDepth

[ECCV 2024] ProDepth: Boosting Self-Supervised Multi-Frame Monocular Depth with Probabilistic Fusion
https://sungmin-woo.github.io/prodepth/
MIT License
22 stars 0 forks source link

The Scale of pose #4

Closed chenlongwhu closed 1 month ago

chenlongwhu commented 1 month ago

Hello, Thank you for your excellent work. I want to know why 

pose = 0.01 * out.view(-1, 6)

what does this code mean? pose[..., :3] is the predicted rotation, pose[...,3:] is the predicted translation. Reducing the translation can control the scope of the depth map, but why reduce the rotation angle?

Sungmin-Woo commented 1 month ago

Hi @chenlongwhu,

The 0.01 scaling is originated from the implementation of MonoDepth2. This scaling factor aids in stabilizing the initial training phase, but it does not seem to be very important overall. Please see monodepth2#260.

I hope this helps :)

chenlongwhu commented 1 month ago

Hi @chenlongwhu,

The 0.01 scaling is originated from the implementation of MonoDepth2. This scaling factor aids in stabilizing the initial training phase, but it does not seem to be very important overall. Please see monodepth2#260.

I hope this helps :)

I got it! Thank you for your reply.