Kaiyang

yes the instructions in readme and the document should have detailed this

you can add the AMSoftmax loss in `torchreid/losses` maybe this could help https://github.com/opencv/openvino_training_extensions/tree/develop/pytorch_toolkit/person_reidentification

https://kaiyangzhou.github.io/deep-person-reid/user_guide#use-torchreid-as-a-feature-extractor-in-your-projects

it's the firewall error caused when trying to download the pretrained model (I assume you're using osnet?) from **google** drive (you know what to do)

is this what you are looking for https://kaiyangzhou.github.io/deep-person-reid/user_guide.html#visualize-activation-maps?

Hi, I have no plan to add Swin to this repo. You are welcome to submit a PR if the model works. :)

the `osnet_ain_*` models are obtained by NAS (please see https://arxiv.org/abs/1910.06827) you can either search a new `osnet_ain_x0_75` or follow the configuration of `osnet_ain_x1_0` to construct the `x0_75` model

if I remember correctly, precise camera labels are unavailable in cuhk03

it's a rare case I'd say I'm not sure whether the problem is caused by colab (probably not) can you make sure cython is indeed not used? this can be...

Sorry for the late reply. > However, the current code cannot achieve this with the default setting. Currently, Feature extractor is initialized with the default device='cuda', it's ambiguous for that...