MhLiao
detecting chinese in images
I have tried to apply the framework of TextBoxes to detect chinese in images,however it doesn't have a good result.Is there any suggestions for improvement?
Yes,I have trained it with the initial weight of the TextBoxes-icdar13.caffemodel on about 8k images with chinese.After about 40000 iterations with base_lr 0.001,the final loss is about 2 and the visual result is bad.Do you have done the same work?
hey @chajbn I'm trying to use TextBoxes in my custom data but i have no idea about how to do it, could you please help me with the steps of training you followed to train chinese dataset ?
@chajbn I also met the same problem. When training my own Chinese scene text image, the final loss is about 1.5. And do you have any suggestion?
For me, I think it is mainly because that the Chinese text in my scene images is too disordered to get a good result quickly. You can have a look at your training dataset.
@https://github.com/dajiangxiaoyan I'm sry to disturb you. I tried to train a new model with my own data , and now i have got the lmdb .But i don't know how to cteate the train.prototxt and what to do next. could u please give me some guidance if you could see the message. And my QQ is 646803015
Have you improved the accuracy?I train the chinese detetion model on my dataset,but the map is only o.23.
