Forked from RACC_NR with add-in support for U-Net.
To enable training in U-Net, simply refer to the DNN_uNet_auto.py script. Type python DNN_uNet_auto.py --help for help. The code is tested under Python 3.5 and should be applicable for version 3.5 and above.
Sample usage:
- To train given a training and a validation dataset: model persistence will be saved to the path specified under the
-dflag.
python DNN_uNet_auto.py train -trgt ../unet/4095_randomized/training/html/ -trn ../unet/4095_noisy_randomized_0p01/training/html/ -vgt ../unet/4095_randomized/validation/html/ -vn ../unet/4095_noisy_randomized_0p01/validation/html/ --ft html -d ../unet/results/
- To predict given a test dataset and a saved model persistence: every single prediction will be saved to the path specified under the
-dflag.
python DNN_uNet_auto.py test -tgt ../unet/4095_randomized/testing/html/ -tn ../unet/4095_noisy_randomized_0p01/testing/html/ --ft html -d ../unet/predictions/train_0p01/test_0p01/html/ -p ../unet/results/html_unet_nopooling.h5
Note: in order to train, one MUST specify both training and testing dataset. As for testing, the flag -p is required to specify where the saved model persistence is. Below is the list of accept parameters:
RACC_NR U-Net script
positional arguments:
<command> 'train' or 'test'
optional arguments:
-h, --help show this help message and exit
-trgt /path/to/dataset/, --traingt /path/to/dataset/
Ground truth folder for the training data
-trn /path/to/dataset/, --trainn /path/to/dataset/
Noisy folder for the training data
-vgt /path/to/dataset/, --valgt /path/to/dataset/
Ground truth folder for the validation data
-vn /path/to/dataset/, --valn /path/to/dataset/
Noisy folder for the validation data
-tgt /path/to/dataset/, --testgt /path/to/dataset/
Ground truth folder for the testing data
-tn /path/to/dataset/, --testn /path/to/dataset/
Noisy folder for the testing data
--ft html Available file type: html, latex, jpeg, pdf
-d /path/to/prediction/, --dest /path/to/prediction/
Path to save the prediction files (at the prediction
time) or model file (at the training time)
-p /path/to/persistence/, --persistence /path/to/persistence/
Path to the saved model file
4095_filetypes_cnn.py is the program for training a file type recognition system of 4095 bit segments of files.
DNN_conv_deconv.py is used to train a convolution-deconvolution network to provide soft information using natural redundancy(NR) which can be supplemented with LPDC decoding for error correction.
DNN_end_to_end_test.py tests an end to end system which does file type recognition based on the system trained by 4095_filetypes_cnn.py,and based on the result, uses the corresponding convolution-deconvolution network trained by DNN_conv_deconv.py. It then stores the soft information so that it is available for the new LDPC decoder
LDPC_ED_BSC.cpp tests the new LDPC decoding aided by NR-based soft information stored by DNN_end_to_end_test.py
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent