search

meijieru / crnn.pytorch

Convolutional recurrent network in pytorch
MIT License
2.38k stars 658 forks source link

Finetuning on my dataset #70

Closed wahrheit-git closed 6 years ago

wahrheit-git commented 6 years ago

I have small dataset of 660 cropped text images. I used the instructions mentioned for training. I changed the display, save and val interval according to my dataset ie. 11 and number of iterations as 750.

I am new to pytorch. Can you suggest me what learning rate shall i use for fine tuning and what was the learning rate that you used? How can i better do the finetuning here like for freezing weights at some layers in pytorch? How and which layers shall i use higher learning rate in pytorch? Which synthetic image generator did you use? Are your results similar to the original CRNN?

meijieru commented 6 years ago

This issue should be asked at PyTorch's forum, so I close it.

wahrheit-git commented 6 years ago

Ok. Can you please tell what learning rate did you use? What specific image generator did you use and how many images you used? How many epoch did you train on?

northeastsquare commented 6 years ago

@kk1153 How about this: https://github.com/JarveeLee/SynthText_Chinese_version/blob/master/gen.py