You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I think the author only provides some general pretrained swin transformer checkpoints. I load the swin_base.pth by network.init_model and missing keys are a lot. I believe the author want users to train their own model using solider. It' ok but still strange not to provide pretrained one to let other people reproduce their results... Or if checkpoints are correct, maybe providing a little more out-of-the-box inference code is a good choice.
Thanks for your attentions. We provide solider models which can be used as a pretrained model to train human parsing task. However, it is general, not specific for human parsing. And we also provide the details of the human parsing training code. Our results can be easily reproduced by running the training code. We believe reproducing the whole training process is more important than just a inference from a provided trained model. Besides, if you have needs of the human parsing trained model, we will release them recently. As the requirements of our company license, it may take a few days. I will catch you up as soon as the model released.
I try to infer human parsing in my data with swin_small pre-trained https://drive.google.com/file/d/1oyEgASqDHc7YUPsQUMxuo2kBZyi2Tzfv/view?usp=share_link
But not succeed on it. This weghts include this task or it is general?
The text was updated successfully, but these errors were encountered: