In the ocrd-train Makefile, here is the code for finetuning
ifdef START_MODEL
$(LAST_CHECKPOINT): unicharset lists $(PROTO_MODEL)
mkdir -p data/checkpoints
lstmtraining
--traineddata $(PROTO_MODEL)
--old_traineddata $(TESSDATA)/$(START_MODEL).traineddata
--continue_from data/$(START_MODEL)/$(START_MODEL).lstm
--net_spec "[1,36,0,1 Ct3,3,16 Mp3,3 Lfys48 Lfx96 Lrx96 Lfx256 O1chead -n1 data/unicharset]"
--model_output data/checkpoints/$(MODEL_NAME)
--learning_rate 20e-4
--train_listfile data/list.train
--eval_listfile data/list.eval
--max_iterations 10000
Why do we need the following line? I thought it was only used in training from scratch.
--net_spec "[1,36,0,1 Ct3,3,16 Mp3,3 Lfys48 Lfx96 Lrx96 Lfx256 O1chead -n1 data/unicharset]" \
Should the learning rate be set lower for fine-tuning? The learning rate for training from scratch is 20e-4, so it would seem that the learning rate for fine-tuning should be significantly lower?
--learning_rate 20e-4 \
--
You received this message because you are subscribed to the Google Groups "tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-oc...@googlegroups.com.
To post to this group, send email to tesser...@googlegroups.com.
Visit this group at https://groups.google.com/group/tesseract-ocr.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/d1c9058e-3885-4ae7-9602-cf9033ccd87f%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.