Closed huangyongbobo closed 2 months ago
We didn't update the model architecture. We just train more iterations on mixed datasets.
Hi @Study-is-happy , thanks for open-sourcing it! What is specifically used for the mixed dataset? I believe this is the model that is refered as "For real-world use cases, we also train the model with a mixed dataset that includes FlyingThings, Sintel, KITTI, HD1K, and VIPER [37]. The pretrained model is available in our repository." in the paper, is that right? How many iterations is used during this step?
Great work! What optimizations have you made to the newly updated neuflow_mixed.pth?