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

princeton-vl / RAFT-Stereo

MIT License

733 stars 140 forks source link

Is it possible to release a model trained on more synthetic data (tartan air/sintel/falling) #51

Open hanskumarKEF opened 2 years ago

lahavlipson commented 2 years ago

Absolutely, I plan on doing this soon

hanskumarKEF commented 2 years ago

Thank you!

hanskumarKEF commented 2 years ago

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

lahavlipson commented 2 years ago

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.