Adam W. Harley

@fding How critical is batch size, in getting it to work? Have you tried with a batch size like 16 or 32, instead of 512? (I am trying it now,...

I have not encountered this! I in fact used multi-gpu training for the models I released. Can you share more about your setup, and your `train.sh`?

OK. So basically I don't know the answer here, but my strategy would be to try to simplify until it works. For a start, how about: comment out the loss...

Ok! 715 is the right number of channels there, so I don't know where 1038 is coming from. Can you add these prints to the `forward` function of `DeltaBlock` in...

Hi @codylcs, sorry, I am updating the repo a bit tonight, and I added a new model ckpt. Can you pull again? I think your issue will disappear.

The first issue there sounds like a bug. Good catch! I will try to fix this soon, and update the results. Separate from this, however, it seems like you're getting...

@Rosie-dala I think I fixed the bug in the badja test script. Check it out: https://github.com/aharley/pips/blob/main/test_on_badja.py#L142-L155

Yes, the released model weights are for S=8. For longer tracking, you need to chain the model over time. There is code for this in [chain_demo.py](https://github.com/aharley/pips/blob/main/chain_demo.py)

In the part where you specify the start locations: https://github.com/aharley/pips/blob/main/demo.py#L31-L37 change it to a dense grid, like this: ``` grid_y, grid_x = utils.basic.meshgrid2d(B, H, W, stack=False, norm=False, device='cuda') xy =...

Tracking all pixels is generally very memory-heavy, and pairing that with the chaining is tricky but doable. The tricky part is: the chaining technique allows each target to choose variable-length...