search

princeton-vl / RAFT-Stereo

MIT License
721 stars 139 forks source link

Finetune on ETH3D #77

Closed Windsrain closed 1 year ago

Windsrain commented 1 year ago

Hello，Could you please tell me how to set num_steps and lr when finetuning on eth3d？I guess that I should use the pretrained model on sceneflow and retrain it with greyscale sceneflow dataset. Is it right?

lahavlipson commented 1 year ago

num_steps=4000 and lr=1e-5

Yes, we finetuned the raftstereo-sceneflow.pth weights on additional greyscale sceneflow images, specifically using --saturation_range 0 0 --img_gamma 0.5 2.0 0.8 1.2