Textboxes++ training - Githubissues

mvoelk / ssd_detectors

SSD-based object and text detection with Keras, SSD, DSOD, TextBoxes, SegLink, TextBoxes++, CRNN

MIT License

302 stars 85 forks source link

Textboxes++ training #34

Closed eugene123tw closed 4 years ago

eugene123tw commented 5 years ago

I'm trying to train my own textboxes++ using the ipython notebook you provided and I noticed an issue:

you use sigmoid (model = TBPP512_dense(softmax=False)) for the background and foreground class confidence instead of using softmax. However, the original SSD training used softmax. And it makes sense since it's not a multi-lable problem. Is there any reason using sigmoid?

mvoelk commented 5 years ago

To my knowledge, the classes in the multiclass case with sigmoid activation are less in conflict with each other then with softmax. In the one class case with sigmoid activation, you actually need only one output (conf_background = 1 - conf_text).

To answer your question, there was no particular reason for me to use sigmoid in this case. So feel free to carry out further experiments... I would be interested in the results :)