taoyang1122
commented
3 years ago Hi, I didn't try the transfer learning experiments. But according to the paper, I think basically you can follow the MOCO codebase to do the finetuning, but you may need to tune the learning rate. BTW, how is the performance if you use MOCO-pretrained model for object detection following the same setting? Is it able to reproduce the results?
Hello, Have you ever try to transfer the simsiam model to object detection task. I have trained a 200-epoch model on ImageNet through simsiam（following the optimal config in paper), Then I take it as pre-trained model of Faster-RCNN and use the transferring config of moco-v2, but I got bad performance. I wonder whether there are different training schedule when transferring self-supervised pre-trained model to object detection task? Hope to get your suggestiones!