Fine-tune on Middlebury and ETH3D

[fuy34]

Hi, Thanks for sharing the code. I found this method also participated in the Robust Vision Challenge and have submissions to both datasets.

May I ask what is the implementation details for that version?

For example, did you also pre-Train the model on Sceneflow and fine-tune on each dataset separately with 300 epochs, or maybe fine-tune on the combined 3 datasets with a certain schedule?

Thanks

[JiaRenChang]

@fuy34 For the Challenge, we fine-tune the model on the combined 3 datasets.