piperod / gammanet_pytorch

gammanet_pytorch

The pipeline can be launched as symple as:

 python run.py config

The config folder contains the example for the meta-configuration that will be used during the experiment in .yaml files:

• datasets: Configuration for the datasets (Test, train and validation). For example inside each .ymal file, the typical configuration metadata would be:

      name: BSDS500_crops
      import_prepath: data.BSDS500_100
      import_class: BSDS500
      images_path: /media/data_cifs/pytorch_projects/datasets/BSDS500_crops/data/images/test
      labels_path: /media/data_cifs/pytorch_projects/datasets/BSDS500_crops/data/groundTruth/test
      transform: 
          Resize:
          size: 320
      input: image
      target: label

• exp: Here the specifications for the experiment is given: Models to be used, logging directories, loss configurations, etc.
• Model: Here the configuration of each of the models that can be used is specified. Where to apply fgru units, attention, saliency, pretrained weights. etc.

Please take a look at each of the files provided for reference.

Organization of the Repo

|____run.py
|____layers
| |____hgru_base.py
| |______init__.py
| |____fgru_base.py
|____config
| |____dataset
| | |____BSDS500_test.yaml
| | |____BSDS500_val.yaml
| | |____BSDS500_100.yaml
| | |____BSDS500_train.yaml
| |____model
| | |____vgg_gammanet.yaml
| | |____sn_hgru.yaml
| | |____vgg_hgru.yaml
| |____exp
| | |____boundary_detection.yaml
|____experiments
| |______init__.py
| |____boundary_detection.py
| |____base.py
|____utils
| |____pt_utils.py
| |______init__.py
| |____py_utils.py
|____models
| |____vgg_16.py
| |____squeezenet.py
| |____vgg_gammanet.py
| |______init__.py
| |____sn_hgru.py
| |____vgg_hgru.py
|____README.md
|____ops
| |____metrics.py
| |____data_tools.py
| |____experiment_tools.pyc
| |______init__.py
| |____optimizers.py
| |______init__.pyc
| |____losses.py
| |____experiment_tools.py
| |____model_tools.py
|____data
| |______init__.py
| |____BSDS500_100.py

Training Example

Single gpu training:

python run.py config/exp/contour_detection

Parallel raining: CUDA_VISIBLE_DEVICES=<gpus> python run.py config/exp/contour_detection parallel=true