Text-Detection-with-FRCN icon indicating copy to clipboard operation
Text-Detection-with-FRCN copied to clipboard

Parameter issues on training coco-data

Open athmey opened this issue 7 years ago • 2 comments

Really appreciate your sharing about the FRCN.

Currently I am following your steps to train the coco-text data.

I use all of the default settings in your script to train the pretrained model, however the result is not so good as yours.(iteration times = 70000, other parameters are the same as that in faster_rcnn_end2end.yml)

Would you please share the essential parameters that you used to train the coco-fine-tuned-model? such as iteration times,learning rate and some other parameters which should be modified. Thank you so much.

athmey avatar Aug 03 '17 01:08 athmey

This is the final result that I got by using the default coco-text database

Evaluating detections Writing text VOC results file VOC07 metric? Yes AP for text = 0.3453 Mean AP = 0.3453

Results:
0.345
0.345

athmey avatar Aug 08 '17 01:08 athmey

hi @athmey , i also use the default settings of frcn, but the pretrain model i use is official VGG16. And do you use full coco-text training data?

jugg1024 avatar Oct 30 '17 15:10 jugg1024