Some doubts about config.py

[dagongji10]

@zhang0jhon Thanks for sharing a nice work! I plan to train my own datasets with AttentionOCR and when I modify config.py, I got some doubts:

1. restore_path: how to use this para? Set it to /path/to/pretrained.ckpt? Need I change starting_epoch?
2. steps_per_epoch: I only have 2 GPUs, so set batch_size=5, and how to change steps_per_epoch? My datasets have 56030 samples, must I set steps_per_epoch=56030/5?
3. warmup_steps: how to use this para? Must it keep same with num_epochs?

[alan-img]

@zhang0jhon Thanks for sharing a nice work! I plan to train my own datasets with AttentionOCR and when I modify config.py, I got some doubts:

1. restore_path: how to use this para? Set it to /path/to/pretrained.ckpt? Need I change starting_epoch?
2. steps_per_epoch: I only have 2 GPUs, so set batch_size=5, and how to change steps_per_epoch? My datasets have 56030 samples, must I set steps_per_epoch=56030/5?
3. warmup_steps: how to use this para? Must it keep same with num_epochs?

your problem I don't know, but I want to know what is label tool of labeling you own datset. because I also want to train in my own dataset.