Valter Hudovernik

Results 8 comments of Valter Hudovernik

Question about how to train mutiple datasets.

I suppose the issue is that AIC has 14 keypoints and thus 14 channels. I think the multi dataset config is intended to train a individual head for each dataset....

[Re] Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations

@rougier No problem. We were under the impression that there was no reproducibility challenge for 2024, so we decided to submit to ReScience C directly.

[Re] Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations

@rougier I understand, no worries. Thank you for moving things along!

[Re] Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations

We decided to submit to [MLRC 2025](https://reproml.org/call_for_papers/).

[Re] Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations

@rougier I am reopening our submission. I added links to the updated article and paper reviews from MLRC.

[Re] Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations

> Ok, are there any public reviews by any chance? Yes, the reviews are available at https://openreview.net/forum?id=GKZ2leags0

[Re] Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations

Thank you for the quick response! > My only two comments concerns the title: [Re] means full replication, [~Re] means partial replication and [¬Re] means not reproducible. Based on your...

make `coalesce` compile friendly

@akihironitta I added a test case, but had to remove the error message as `torch._check` requires the message to be callable and passing a lambda function causes a graph break...