search

HKUST-Aerial-Robotics / MVDepthNet

This repository provides PyTorch implementation for 3DV 2018 paper "MVDepthNet: real-time multiview depth estimation neural network"
GNU General Public License v3.0
309 stars 72 forks source link

more image pose pairs #10

Closed boni-hu closed 5 years ago

boni-hu commented 5 years ago

Dear WANG-KX, Thanks for you great project, I have used two image pose pairs to predict depth, the result is not very good, now I want to know how to use more than two image pose pairs to predict depth. Thanks for you reply.

Sincerely, HuBoni

WANG-KX commented 5 years ago

It's pretty simple. Just take the average of [the cost volume] from different pairs (make sure the left images are the same)(https://github.com/HKUST-Aerial-Robotics/MVDepthNet/blob/7a13790369559a365ee0fd914f12cfb3524c759b/depthNet_model.py#L139) and put it into the net.

boni-hu commented 5 years ago

Dear, Thank you very much. I used ten measurement frames and one reference frame to compute cost volume,and predict depth ,as result the depth slightly better than two pictures.Should there be more pictures to compute costvolume? Sincerely, Huboni

WANG-KX commented 5 years ago

Dear, I think 10 images are far more than 'enough'. As far as I think, high-quality images with good pose estimations are equally important. If you are in forward/backward movements, more images can help. If you have left/right stereo-like configuration, ten images are enough.

boni-hu commented 5 years ago

Okay,thank you very much!

Huboni