Training pdNet

This work is under review for Special issue for Pattern Recognition letters on Graphonomics.

Currently the network can be trained on synthetic and DIBCO dataset:

Datasets	Input Resolution	Output Resolution^	# of classes
Synthetic	128x256	16x32	2
DIBCO	128x256	16x32	2

^ is the encoder output resolution; decoder output resolution is the same as that of the input image. The datasets are avaialble here

Files/folders and their usage:

• run.lua : main file
• opts.lua : contains all the input options used by the tranining script
• data : data loaders for loading datasets
• models : all the model architectures are defined here
• train.lua : loading of models and error calculation
• test.lua : calculate testing error and save confusion matrices

Example command for testing the code on DIBCO:

Training encoder Synthetic data:

th run.lua --dataset sn --year 0000 --datapath ../datasets/xSynthetic/128x256 --cachepath ../resources/Dibco/128x256/cache/encoder --model models/encoder.lua --save ../resources/Dibco/128x256/trained/encoder --imHeight 128 --imWidth 256 --labelHeight 16 --labelWidth 32 --batchSize 30 --maxepoch 20

Training decoder Synthetic data:

th run.lua --dataset sn --year 0000 --datapath ../datasets/xSynthetic/128x256 --cachepath ../resources/Dibco/128x256/cache/decoder --model models/decoder.lua --save ../resources/Dibco/128x256/trained/decoder --imHeight 128 --imWidth 256 --labelHeight 128 --labelWidth 256 --batchSize 30 --maxepoch 20 --CNNModel ../resources/Dibco/128x256/trained/encoder

Training encoder on DIBCO data for year 2009 using pre-trained encoder model:

th run.lua --dataset db --year 2009 --datapath ../datasets/xDibco/128x256 --cachepath ../resources/xDibco/128x256/cache/encoder --model models/encoder.lua --save ../resources/xDibco/128x256/trained/encoder --imHeight 128 --imWidth 256 --labelHeight 16 --labelWidth 32 --batchSize 30 --maxepoch 20 --ptModel ../resources/xDibco/128x256/trained/encoder/model-0.net

Training decoder on DIBCO data for year 2009 using pre-trained decoder model:

th run.lua --dataset db --year 2009 --datapath ../datasets/xDibco/128x256 --cachepath ../resources/xDibco/128x256/cache/decoder --model models/decoder.lua --save ../resources/xDibco/128x256/trained/decoder --imHeight 128 --imWidth 256 --labelHeight 128 --labelWidth 256 --batchSize 30 --maxepoch 20 --CNNModel ../resources/xDibco/128x256/trained/encoder --ptModel ../resources/xDibco/128x256/trained/decoder/model-0.net

Training pdNet on DIBCO data for year 2009 using trained decoder model:

th run.lua --dataset db --year 2009 --datapath ../datasets/xDibco/128x256 --cachepath ../resources/xDibco/128x256/cache/decoder --model models/pdNet.lua --save ../resources/xDibco/128x256/trained/pdNet --imHeight 128 --imWidth 256 --labelHeight 128 --labelWidth 256 --batchSize 20 --CNNModel ../resources/xDibco/128x256/trained/decoder --maxepoch 20	

Test the output on DIBCO 2009 dataset:

cd ../visualize
th demo_PDdibco.lua -i ../datasets/xDibco/128x256/valid -d ../resources/xDibco/128x256/trained/ -m pdNet --year 2009 --dataset db -o ../datasets/xDibco/xOutput_pdnet_128x256 --devID 2