Closed zydou closed 7 years ago
The best checkpoints rarely occurred in the first few epochs. So adding this option will reduce the trainning time. For example:
python examples/triplet_loss.py -d market1501 -a resnet50 --combine-trainval --logs-dir examples/logs/triplet-loss/market1501-resnet50
and
python examples/triplet_loss.py --start_save 130 -d market1501 -a resnet50 --combine-trainval --logs-dir examples/logs/triplet-loss/market1501-resnet50
gives the same results. But the latter is faster.
The best checkpoints rarely occurred in the first few epochs. So adding this option will reduce the trainning time. For example:
and
gives the same results. But the latter is faster.