Closed helloyuning closed 1 year ago
Hi, thanks for your interest in our work. Can you specify your question 1? I don't fully understand it. Do you refer to the trained network or the detailed implementation of the models? As for question 2, the pruning ratio is an empirical number. It is a trade-off between efficiency and accuracy. As for COCO, we find that a prune ratio of 0.7 would be fine. You can see the ablation study in our paper for more details.
I would like to ask where the specific code implementation of prune step is, It seems that the code of the model does not include this part.
thanks
This code does include the pruning step. As we only prune the tokens, rather than the network weights, we only need to make a slight change on TokenPose. You can find the detailed token pruning part at this code.
Tanks for your greate work and sharing.
I have some questions:
Tanks , yuning