Created by: hshen14

I followed the comments in https://github.com/PaddlePaddle/Paddle/issues/13574#issuecomment-424587581 to reproduce the accuracy of CRNN-CTC after a minor fixing below:

diff --git a/fluid/ocr_recognition/eval.py b/fluid/ocr_recognition/eval.py index 666803d..87f4f9c 100644 --- a/fluid/ocr_recognition/eval.py +++ b/fluid/ocr_recognition/eval.py @@ -31,7 +31,7 @@ def evaluate(args): num_classes = data_reader.num_classes() data_shape = data_reader.data_shape() # define network

• evaluator, cost = eval(data_shape, num_classes)

• evaluator, cost = eval(data_shape, num_classes, True if args.use_gpu else False)

I got the below metrics: Read 2000 samples; avg_distance: [0.08513155]; avg_seq_error: [0.2255]

Can we understand the output is SOTA accuracy? How can I understand the meaning of the avg_distance and avg_seq_error? Is there any formula behind? Thanks.