Closed ReverseSystem001 closed 4 years ago
Yes you are right
thanks for your quickly replay. can you tell me the training details of Tiny NASNet-A? the epoch, iterations,and learning rate.
Please see https://arxiv.org/pdf/1809.01407.pdf Appendix A1 and A3. For Tiny NASNet-A, for supervised initialization (1/11 data), lr is 0.1, batch size is 256, M is 50; for joint training (full data), please see A3
thanks.
Does the architecture of Tiny NASNet-A in https://github.com/XiaohangZhan/face_recognition_framework/blob/master/models/backbones/nasnet.py, and the NASNetASmall in the git link is the same with Tiny NASNet-A, right?