Kenneth Kong

https://github.com/Alibaba-MIIL/ASL/blob/7068ff007ac5f56a0ae1794b61a7571c3e4561ec/src/loss_functions/losses.py#L107

I change the centerness head to regression head, but found it isn't work in my experiment, is someone else did this experiment?

As the title expressed, thank u a lot!

- here is where it use drop path: https://github.com/microsoft/Swin-Transformer/blob/main/models/swin_transformer.py#L217 - and here is timm implementation: https://github.com/huggingface/pytorch-image-models/blob/main/timm/layers/drop.py#L173 Should it set scale_by_keep=False explicitly?

paper中说：in comparison with our baseline, translation of the prompts provided by OpenAI CLIP. The baseline with around 90 prompts performs the best on average 但是这个templatehttps://github.com/OFA-Sys/Chinese-CLIP/blob/master/cn_clip/eval/cvinw_zeroshot_templates.py#L6，实际上180+，这个你们后期调过么？我看部分数据集（包括cifar100）也使用的这个template（而不是使用数据集自身对应的template)，是对比过么？