Could you provide us training configuration more precisely?

valencebond / Rethinking_of_PAR

Pytorch Pedestrian Attribute Recognition: A strong PyTorch baseline for pedestrian attribute recognition and multi-label classification.

181 stars 37 forks source link

Could you provide us training configuration more precisely? #4

Closed macarize closed 2 years ago

macarize commented 2 years ago

I trained your model with swin_s for PA100K but there was noticable gap between my experiment(79.48) and your reported performance(80.35) I assume that configuration i used for training is different from your settings. Could you provide us more precise configuration for swin_s (PA100K)? many thanks.

valencebond commented 2 years ago

hi @macarize I train the swin_s model by replacing the BACKBONE.TYPE as "swin_s" in pa100k.yaml. Other configurations all remain the same. What is the epoch of your performance?