Kanji recognition assignment. This code can run in 2 modes : Test (using already trained model) and train mode. Number of categories considered: 100 Standard image size : [64,64,1]
Test mode: The runs from a file "kanji_rec_test.py". The model uses for recognition is stored in "model_kanj_irec64.ckpt". usage: $ python kanji_rec_test.py ....
Expected output: labelID : xxxx ; Recognized Kanji : X labelID : xxxx ; Recognized Kanji : X ....
Train mode:
The training code is in the file: kanji_rec_train.py It needs testDB files to be present in the directory ./dest64/ The files that are needed are
- Training Data file : 35000 samples
- Training Label file
- Test data file: 10000 samples
- Test data label file
- validation data file: 5000 samples
- validation label file
These files are generated by processing the image DB offline. These files are in the format: Data file : [N, IMAGE_SIZE, IMAGE_SIZE, NUM_CHANNELS]/ N: number of images in a DB IMAGE_SIZE: resized image size : 64x64 NUM_CHANNEL: 1 (considered gray image)
Label file : 1xN D array