Closed linchuanze closed 2 years ago
Hi, for RRT you can follow the command
python train.py [...retrained model arguments...] --retrain_fc --pretrained <path_to_pretrained_ckpt>
For RRT, the hyper-parameter settings remain the same between the first and the second training stage, though one usually trains for fewer epochs as only last layer is trained (e.g., 10/30 epochs for age datasets). See examples & descriptions from, for example, AgeDB-DIR.
Do i need set a few epoch to traing the last regressor.As shown in the figure below, what causes it