Attention_OCR

CNN+Attention+Seq2Seq

  • The model and its tensor transformation are shown in the figure below

  • It is necessary ch_ train and ch_ test the picture address format of test text to its own file path format

  • There is a missing data picture in the data set originally given in the test set, and there is an empty picture in the picture data set

The path in the text is as follows

/mnt/disk2/std2021/hejiabang-data/OCR/attention_img/AttentionData/59041171_106970752.jpg 项链付出了十年的苦役
/mnt/disk2/std2021/hejiabang-data/OCR/attention_img/AttentionData/38115031_1485663711.jpg 。直到台“国防部长”
/mnt/disk2/std2021/hejiabang-data/OCR/attention_img/AttentionData/22905328_1196841476.jpg 有惊无险地以21比1
/mnt/disk2/std2021/hejiabang-data/OCR/attention_img/AttentionData/41681796_2460379288.jpg 尼在门前两米处上演“
....

The training results are as follows

GitHub

View Github