search

detectRecog / CCPD

[ECCV 2018] CCPD: a diverse and well-annotated dataset for license plate detection and recognition
MIT License
2.22k stars 566 forks source link

Has anyone achieved the accuracy in the paper through the recurrence model? #33

Open SaulZhang opened 5 years ago

SaulZhang commented 5 years ago

Can the author publish a detailed data set partitioning method or publish a final prediction model?

SaulZhang commented 5 years ago

image This is the result of my experiment. I divide the dataset with train:test=100k:100k, the same as the paper. But it seems that the performance is pretty poorer than the paper's.

SaulZhang commented 5 years ago

However, it is worth noting that although my experimental results are generally worse than the original one, the above items in data set "Weather" are superior to the original ones. Why is this happening?

Doylegyf commented 5 years ago

My result is also poor

zhqx511 commented 5 years ago

so do I

yangdaxia6 commented 5 years ago

image This is the result of my experiment. I divide the dataset with train:test=100k:100k, the same as the paper. But it seems that the performance is pretty poorer than the paper's.

Did you just divided the ccpd_base into 100k：100k and not diveded other subdata? And your recognition precision is larger than detection precision, how did you calculate the recognition and detection precision