Jian
Jian
if I understand right, `relative_position_bias_table` only depends on window size ( 7 for 224 in swin_tiny), so `relative_position_bias_table` do not need to change. `attn_mask` is not learnable parameter, which can...
hai @JiyueWang , i can just achieve 65.67 accuracy with resnet50 under torch1.4. is there any thing i need to change except you mentioned above? thanks~
same problem. I can install it without error. But when `from inplace_abn import InPlaceABN`, the same error occurs with python 3.7 pytorch1.7 gcc7.5 > inplace_abn-1.1.1.dev2+g938ffd2-py3.7-linux-x86_64.egg/inplace_abn/_backend.cpython-37m-x86_64-linux-gnu.so: undefined symbol: cudaSetupArgument
> same problem. I can install it without error. But when `from inplace_abn import InPlaceABN`, the same error occurs with python 3.7 pytorch1.7 gcc7.5 > > > inplace_abn-1.1.1.dev2+g938ffd2-py3.7-linux-x86_64.egg/inplace_abn/_backend.cpython-37m-x86_64-linux-gnu.so: undefined symbol:...
hi @Dorniwang , if eq2 is wrong, so all the following formulation is wrong ? or Is there any influence on following formulation?
Hi, did you find some configs? thanks!
@wduo @abhigoku10 Pretrained models will be provided in the next few weeks. What is demo.py used for?
@abhigoku10 we have already provided the infer.py. You can just replace the dataloader with a single image.
@abhigoku10 We are too busy right now. You can train the model by yourself, it takes less than 12 hours.
the names of 35 selected attributes are given in https://github.com/valencebond/Rethinking_of_PAR/tree/master/dataset/pedes_attr/annotation.md