nicolas-dufour

Incompatibility with torch.fx

6

**Describe the bug** When trying to call torchvision.models.feature_extraction.get_graph_node_names on a model that has an einops operation, the operation fail and gets the following error Traceback (most recent call last): File...

The best theme for jupyter lab: https://github.com/mohirio/jupyterlab-horizon-theme

Computation of the loss reweighting

1

Hey, I've noticed that the weight reweighting that is used in the code doesn't match the papers equation Indeed, here, the formula is BxHxW/(num_pixels_of_class_i*num_of_non_zeros_classes_in_batch) whereas the paper showcase the following...

Training Speed

1

Hi! The authors from the paper said that they needed to use a custom conda kernel to get acceptable performance (They say it's 20x faster than with pytorch primitives. I...

Dataset availability

3

Hi, I'm trying to retrieve the LIP dataset but the link provided seems to be dead. Do you happen to provide a backup solution? Thanks

[CLI]: When training in offline mode, wandb tries to connect after a while a crashes the run

3

### Describe the bug Hi, i'm training using pytorch lightning wrapper for wandb using offline=True. After 4h of training with no problems, my code get holded indefinitly with the following...

Details on the training procedure

2

Hi, Could you provide details on the training procedure of this model? How much steps were used on the stage 1 of the model (autoencoder)? On which dataset? Is it...

Discrepancy with CMMD paper

2

Hi, I think there is a bug in the CMMD implementation with respect to the paper. In the paper, the MMD distance is  However, looking at the code, $K_{X,X}$...

oscillations in the loss

2

Hi! I see in the loss for cc3m model that there are oscillations in the loss. When using the same webdataset framework I also have oscillations as well. In my...

Hi, would it be possible to have the original alt-text before anonymisation? There is quite a lot of noise in the anonymisation process, and it would be helpful to get...