search

jzbontar / mc-cnn

Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches
BSD 2-Clause "Simplified" License
707 stars 232 forks source link

a small question regarding adcensus.cu #31

Open laoreja opened 7 years ago

laoreja commented 7 years ago

I was looking for the docs of adcensus.cu, but cannot find one. And later, I noticed your adcensus repo.

So is the adcensus.cu a part of your adcensus repo?

laoreja commented 7 years ago

Because through googling 'adcensus', I got lots of results, so I thought it's a library from somewhere.

jzbontar commented 7 years ago

I was looking for the docs of adcensus.cu, but cannot find one.

There are no docs for adcensus.cu, sorry. The file contains the CUDA side of the mc-cnn implementation. Is there anything specific you would like to know about it?

And later, I noticed your adcensus repo. So is the adcensus.cu a part of your adcensus repo?

There is no connection, other than the name, between adcensus.cu in this repo and my other repository https://github.com/jzbontar/adcensus.

As a short historical note, when I started working on stereo, I wanted to reimplement a good baseline method first. I decided to go for the ad-census paper, because it was relatively easy to understand and performed well on the Middlebury dataset. This is the story behind https://github.com/jzbontar/adcensus. Afterwords, I played around with the code, trying to improve it and after a few months, the file adcensus.cu evolved into the mc-cnn architecture that you can find in this repo. I never got around to changing the name adcensus.cu to mc_cnn.cu, which would be more appropriate.