BCER (best character rate) is automatically picked by tesseract from all list of character rates errors (CER).
On Thursday, November 23, 2023 at 12:34:40 PM UTC+3 Des Bw wrote: > I think they are abbreviations: > best char error =BCER > character error = CER > > There is no signs to tell if the model is overfit. I know no diagnostics > for that. For fine-tuning, running iterations higher than 400 is always > problematic because it destroys the base model. > > - So, the common strategy is to increase your data; and run just 300 > iterations. The BCER is not that important in that case. > But, for training from scratch or from layer (network), you should try to > get the BCER (error rate) as low as possible. Overfitting happens when the > data is too small, and the iterations are too many. From my experience, > running 2-5 epochs seems to generate good results. But, I have seen > experienced guys training for hundred even thousands of epochs. > > > On Thursday, November 23, 2023 at 12:28:35 PM UTC+3 [email protected] > wrote: > >> Alright, >> >> this might be a litte bit of a dump question but where exactly can I see >> the CER? >> >> 2 Percent improvement time=56, best error was 12.49 @ 8294 >> At iteration 8350/10000/10000, Mean rms=2.701%, delta=2.491%, char >> train=10.385%, word train=24.4%, skip ratio=0%, New best char error = >> 10.385 wrote best >> model:data/Common_num/checkpoints/Common_num10.385_8350.checkpoint wrote >> checkpoint. >> >> Is it the "best char error"? Where do I have to look to find CER? Is the >> CER in the above example? >> >> Also what are signs that my model is overfitted? Is there any possibility >> recognicing this in the above statement? >> >> -- 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 [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/94c279c1-5e87-4c80-948a-b0f87e64c5b6n%40googlegroups.com.
