search

clovaai / deep-text-recognition-benchmark

Text recognition (optical character recognition) with deep learning methods, ICCV 2019
Apache License 2.0
3.72k stars 1.09k forks source link

When ground truth is incorrect but, prediction is well-predicted. #394

Open ghost opened 1 year ago

ghost commented 1 year ago

Hello, I'm using this code right now for License plate recognition.

But, Ground Truth is not correct for training/validation case.

My option is Following.

------------ Options ------------- exp_name: exp_003 train_data: dataset/LP_train3 valid_data: dataset/LP_val2 manualSeed: 1111 workers: 0 batch_size: 128 num_iter: 60000 valInterval: 1000 saved_model: exp_002.pth FT: True adam: False lr: 1 beta1: 0.9 rho: 0.95 eps: 1e-08 grad_clip: 5 baiduCTC: False select_data: ['/'] batch_ratio: ['1'] total_data_usage_ratio: 1.0 batch_max_length: 25 imgH: 32 imgW: 100 rgb: False character: (my Characters) sensitive: False PAD: False data_filtering_off: True Transformation: TPS FeatureExtraction: ResNet SequenceModeling: BiLSTM Prediction: Attn num_fiducial: 20 input_channel: 1 output_channel: 512 hidden_size: 256 num_gpu: 1 num_class: 51

Such case is following

Ground Truth | Prediction | Confidence Score & T/F

  1. | 112916. | 0.9918 True

There is no latter on gt. But, prediction is good.

my gt file has correct data and exact name of LP.

How can i handle this?

Please give me some idea.

thanks.

yasaslive commented 1 year ago

What is your character set? If its is upper then you need to add lower as well

ghost commented 12 months ago

What is your character set? If its is upper then you need to add lower as well

I see.. I check for English case. Lower case is working.

But, For korean case, there is no upper :(

Anyway, Thanks for the information for lower / upper case 👍

yasaslive commented 12 months ago

You can remove the .lower() function in dataset.py