Closed Shreeshrii closed 3 years ago
CER Data was extracted as follows:
grep 'Eval Char' /home/ubuntu/tess5training-iast/LAYER.log | sed -e 's/^.*[0-9]At iteration //' | \sed -e 's/,.* Eval Char error rate=/\t/' | sed -e 's/, Word.*$//' | sed -e 's/^/\t\t/'> plot-eval.txt
grep 'best model' /home/ubuntu/tess5training-iast/LAYER.log | sed -e 's/^.*\///' | sed -e 's/\.checkpoint.*$//' | sed -e 's/_/\t/g' | sed -e 's/\(.*\)\t\(.*\)/\1/' > plot-best.txt
grep 'At iteration' /home/ubuntu/tess5training-iast/LAYER.log | sed -e '/^Sub/d' | sed -e '/^Update/d' | sed -e 's/At iteration \([0-9]*\).*char train=/\t\t\1\t\t/' | sed -e 's/%, word.*$//' > plot-iteration.txt
sed 'N;s/\nAt iteration 0, stage 0, /At iteration 0, stage 0, /;P;D' /home/ubuntu/tess5training-iast/CHECKeval.test.log | grep 'Eval Char' | sed -e 's/.checkpoint.*Eval Char error rate=/\t\t\t/' | sed -e 's/, Word.*$//' | sed -e 's/\(^.*\)_\(.*\)_\(.*\)\t/\1\t\t\2\t\t\t/g' > plot-validation.txt
cat plot-header.txt plot-validation.txt plot-best.txt plot-eval.txt plot-iteration.txt > plot_cer.csv
python plot_cer.py
@kba I have added the info about how to run this in README.
|tee -a LAYER.log
does not capture the output from lstmtraining. I use nohup
for capturing all output.
LAYER training is NOT currently supported by Makefile. I had used logfile from an independently run `replace top layer training'.
I used the sample set provided in this repo and ran training from scratch as well as with START-MODEL. Those plots look a bit different. See below.
as suggested in https://github.com/tesseract-ocr/tesstrain/issues/200#issuecomment-719579821