Compression result of Faster Rcnn on COCO
Edwardwaw opened this issue · comments
thx for your great job! i wonder about the compression result of Faster Rcnn on larger dataset like coco, which isn't seen in your paper. could you share some information about this matter?
thx for your great job! i wonder about the compression result of Faster Rcnn on larger dataset like coco, which isn't seen in your paper. could you share some information about this matter?
Hi, thx for your kind words! The main reason why we did not provide the compression results of Faster RCNN on COCO is that the code we adopted for the pruning of Faster RCNN only supports single GPU training. Therefore, it would be time-consuming for us to fine-tune the model on COCO. And considering there was little pruning research conducted on the VGG16-based Faster R-CNN to our knowledge at that time, we decided to skip this experiment and validate the effectiveness of our method on more cutting-edge experiments that support multi-GPU training on a large benchmark. Hope the reply can address your question.