Some bad results on kid-running case

[STLAND-admin]

Hi zhengqi！

I tried to train the kid-running case but get some bad results, the only change is the chunk_size and N_rand to fit my GPU memory.

So I am also working on the problem and want to know if you can release the Kid-running checkpoint, I didn't find it with those links you offered.

update-------------------------------------------------------------------------------------------------------------
Well, I tried to increase the size of the above two parameters. The results seem much better. However, there are still some artifacts in some view.

Do you have the same problem in your checkpoints or training process? If not, would you please offer the checkpoints?

Thanks!

[zhengqili]

Hi. I observed such artifacts before for some of videos I tried. Usually it's due to insufficient initialization, bad source view feature selections, or small sampling pixel size. For this example, using default hyper-parameters is less likely to produce such artifacts. However, I can share the checkpoints I trained if you are interested.

[STLAND-admin]

Yes, it will be nice if you can share the checkpoints, my email is [email protected]