Soufiane Belharbi

icon indicating a website link https://sbelharbi.github.io

icon location Montréal, Québec icon location Post-doc at LIVIA Lab. ÉTS Montreal, in collab. with McCaffrey Lab. /GCRC McGill. Training neural networks with weak supervision.

Results 5 issues of Soufiane Belharbi

update pl.p: table creation

create table directly in device instead on cpu + transfer. thanks

what is aux loss mentioned in the paper?

hi, there was a mention of an auxiliary loss 'aux loss' in the [paper](https://openaccess.thecvf.com/content/CVPR2021/papers/Wei_Shallow_Feature_Matters_for_Weakly_Supervised_Object_Localization_CVPR_2021_paper.pdf), but it doesnt seem to be explained what it is or i may have missed it....

weights' links are not working.

2 comment

hi, the provided links of the weights in the table (readme) are all not working. i tested all of them. ![image](https://user-images.githubusercontent.com/23446793/224879559-3c7e0ceb-e8fc-4139-9a74-8453666e6062.png) i am interested in pytroch version, (resnet50 in particular)....

Master

action units maps extraction

3 comment

hi, can you please confirm that `gh2` is the attention maps? https://github.com/rakutentech/FAU_CVPR2021/blob/0bfb778526908f36b6136e836d8b382877bacfa4/inference.py#L53 it is of size `(batch_size, 12, 12, number_action_units)`. here, `number_action_units=12`. attention maps are output of the arrow in...