Open HuangLian126 opened 4 years ago
Method for training deep unsupervised sailency detection model. Training module implements 4 training methodology
- Real: The backbone network is trainined with Ground Truth object, loss = mean((pred_y_i, y_gt_i))
- Noise: The backbone network is trained using losses on all the labels of unsupersived methods, loss = mean((pred_y_i, unup_y_i_m))
- Avg: The backbone network is trained using avg of the all the unsupervised methods as ground truth, loss = mean((pred_y_i, mean(unup_y_i_m)))
- Full: The backbone network as well as the noise module is trained, the training proceeds as follows
You can refer to https://github.com/kris-singh/Deep-Unsupervised-Saliency-Detection/blob/master/src/deep_unsup_sd.py for more details.
@Patrickctyyx Thanks for your reply. I'm training, hhh.
Thanks for your implmentation ! Recently, I am interested in the unsupervised learning, this implmentation can help me. Here, I don not know the meaning of "Exp Name FULL". Can you explain the Full Training?