Thank you for your open source work on this project. I use your open source code and default configuration to train in somethins-v1 data set, and the accuracy is always about 10 points lower than the accuracy in the paper. May I ask why there is such a big gap?
Thank you for your open source work on this project. I use your open source code and default configuration to train in somethins-v1 data set, and the accuracy is always about 10 points lower than the accuracy in the paper. May I ask why there is such a big gap?