princeton-vl/RAFT-Stereo

Is it possible to release a model trained on more synthetic data (tartan air/sintel/falling)

hanskumarKEF opened this issue · 4 comments

Is it possible to release a model trained on more synthetic data (tartan air/sintel/falling)

Absolutely, I plan on doing this soon

Thank you!

Was wondering if these models were still going to be released?

Sorry for the delay, I trained a RAFT-Stereo model on some existing synthetic stereo datasets newer/larger than those listed in this paper, although the performance benefit was marginal, if not worse.

Instead we'll release a better model trained on non-public data, hopefully soon. I'll make sure to update this issue when that happens.