search

mrharicot / monodepth

Unsupervised single image depth prediction with CNNs
Other
2.22k stars 631 forks source link

Newer versions of code #174

Closed varun19299 closed 6 years ago

varun19299 commented 6 years ago

Thank you for the tf implementation, we've found it quite useful.

However, the repo seems not updated from changes that have occurred since Nov 2017 intf (better IO pipelines, Estimators, visualisation).

We've used these modifications to distribute training and add more feature extractors.

Are there any plans to continue support for this repo? In which case, would a pull request be accepted?

mrharicot commented 6 years ago

Hi,

I will still support this repo for bug-fixing only but won't be updating it to the more recent API to simplify maintenance. At some point in the future we'll release the code of our new method as it's simpler to modify and produces significantly better results in less time.

varun19299 commented 6 years ago

Nice, does your research include (chronologically speaking) https://arxiv.org/pdf/1802.05522.pdf. This was an unsupervised Learning of Depth and Ego-Motion from Monocular Video Using 3D Geometric Constraints, and they had cited your paper too.

The whole of monocular depth estimation seems to be a rapidly changing area, rather becoming hard to keep track of now :P.

mrharicot commented 6 years ago

monodepth 2 is more recent than this paper. We did compare with it as well as with all the cvpr 2018 papers doing depth estimation, have a look at it to find all the recent papers.

monocular depth estimation is indeed a bit of a wild west at the moment :)