Closed liuzhidemaomao closed 2 years ago
It means that the backbone networks are pretrained on ImageNet 1k/22k, which is a common practice in semantic segmentation.
It means that the backbone networks are pretrained on ImageNet 1k/22k, which is a common practice in semantic segmentation.
Thank you for your reply.
In ISNet, the models are fine-tuned for 160K iterations on ADE20K and LIP datasets. So do you pretrained the models on other datasets?