Media-Smart / vedastr

A scene text recognition toolbox based on PyTorch
Apache License 2.0
534 stars 100 forks source link

Can't reach the score of cstr in paper #70

Open w867066886 opened 3 years ago

w867066886 commented 3 years ago

I train it in three RTX3090 and get the average score 86. Are there any other details that have not been revealed

ChaseMonsterAway commented 3 years ago

Hi, Did you modify other parameters in configure file except gpu_nums?

w867066886 commented 3 years ago

I don't modified anything except the gpu_nums and the dataset path

ChaseMonsterAway commented 3 years ago

Can you upload you training log?

w867066886 commented 3 years ago

Sorry, I deleted the training log by mistake. I will train again and then give feedback

ChaseMonsterAway commented 3 years ago

Sorry, I deleted the training log by mistake. I will train again and then give feedback

You'd better follow the steps in README. For example, you should use the SyntheText released by ASTER since we train the model using it.

w867066886 commented 3 years ago

Thanks. By the way, MJ ,validation and evaluation datasets is from deep-text-recognition-benchmark, ST is from aster, right?

ChaseMonsterAway commented 3 years ago

Thanks. By the way, MJ ,validation and evaluation datasets is from deep-text-recognition-benchmark, ST is from aster, right?

Yes.