pytorch-metric-learning

pytorch-metric-learning copied to clipboard

distributed training bug fix, to maintain same order for mem queue fo…

16

…r all workers

zhaoyuac09

Since metric learning is very useful for ranking tasks, I'd like to suggest including the [ApproxNDCGLoss](https://www.tensorflow.org/ranking/api_docs/python/tfr/keras/losses/ApproxNDCGLoss) and other ranking losses such as [L2R losses](https://pytorchltr.readthedocs.io/en/stable/loss.html) using the PML approach.

celsofranssa

I implemented the [Multi-SupConLoss](https://sslneurips22.github.io/paper_pdfs/paper_8.pdf) as mentioned in issue #658. However, this implementation rewrote a lot of functions because many functions conflicted with the multi-label task such as [check_shapes](https://github.com/KevinMusgrave/pytorch-metric-learning/blob/ac607007dc62666f9de850cd5b8e5694ff0da1c2/src/pytorch_metric_learning/utils/common_functions.py#L389) and [get_matches_and_diffs](https://github.com/KevinMusgrave/pytorch-metric-learning/blob/ac607007dc62666f9de850cd5b8e5694ff0da1c2/src/pytorch_metric_learning/utils/loss_and_miner_utils.py#L31)....

QibinLiang

Default normalization in distances is counterintuitive (or wrong)

1

Not a real bug, and maybe it's just personal preference, but I feel like the normalization in several distances is counterintuitive. For example, the documentation for `CosineSimilarity`, says > This...

mcschmitz

Hello , Is this package able to train a multi-label dataset with one-hot encoding?

2

It is a great package that improves my efficiency , When I test cifar10, for the one-hot label , I can use ```python label = torch.argmax(label,dim=1) ``` to transform one-hot...

Mahiro2211

Multi-label version of SupConLoss

2

This [paper](https://sslneurips22.github.io/paper_pdfs/paper_8.pdf) introduced a new loss function called `MultiSupCon` that allows us to gain knowledge about the degree of label overlap between pairs of samples. I was just wondering if...

penguinwang96825

I have put my thought in refactoring reducers and mixins. In some cases i have updated the tests relative to new functionality that could be added easily to the reducers....

domenicoMuscill0

Fixed issues #491, #430 and #602

3

I have tried to solve these issues. All the pre-existing tests work fine, but i did not create new ones since i have just refactored the same code. I have...

domenicoMuscill0

As suggested and implemented in pull request #635 by @domenicoMuscill0. I'm removing this feature from that PR because it's already a large PR. Here's the implementation by @domenicoMuscill0: ```python def...

KevinMusgrave

See #429

KevinMusgrave