LightViT
LightViT copied to clipboard
Official implementation for paper "LightViT: Towards Light-Weight Convolution-Free Vision Transformers"
您好,请问下为什么提供的训练好的模型是ckpt文件,不是pth文件?是用tf训练的吗?并且在轻量化上,LightViT可能主要对比的结构是transformer一类,而不是mobilenet等常用的轻量级网络?因为自己训练出的LightViT-small模型有200多M。谢谢~
Hi, I am trying to train lightvit tiny model, but I faced this issue. What is the `mc module` that you mentioned in the code? > bash tools/dist_train.sh 2 configs/strategies/lightvit/config.yaml...
pretrained_weights_path = 'lightvit_tiny_78.7.ckpt' pretrained_state_dict = torch.load(pretrained_weights_path) lightvit = lightvit_tiny(pretrained=False) lightvit.load_state_dict(pretrained_state_dict) I use about code to use pretrained lightvit_tiny model, but it's useless, how can i pretrained lightvit_tiny model correctly?
test.py: error: unrecognized arguments: --local_rank=0