BioNet

Deep Convolutional Neural Networks with bio-inspired filters.

1. Clone this BioNet repository
2. Clone the CIFAR-10G generalisation test set
3. Set your project_dir in the notebook and pass your data_dir (ln -s /shared/data/ data) which contains the image sets

Expected directory structure

.
├── bionet
│ ├── config.py
│ ├── explain.py
│ ├── __init__.py
│ ├── plots.py
│ └── preparation.py
├── data
│ ├── CIFAR-10G
│ ├── ecoset
│ └── ecoset-cifar10
├── logs
├── models
├── notebooks
├── results
├── scripts
├── model.py
└── README.md

Training and testing the model

The main script to handle training and testing is model.py in the project's root directory. This script is called to both train and test the models. If saved weights files are found in the models directory, training will be skipped (unless the clean flag is passed) and the code will proceed to testing.

Arguments and usage

usage: model.py [-h] [--convolution CONVOLUTION] [--base BASE] [--pretrain]
[--architecture ARCHITECTURE] [--interpolation INTERPOLATION]
[--optimizer {SGD,RMSprop,Adagrad,Adadelta,Adam,Adamax,Nadam}]
[--lr LR] [--decay DECAY] [--use_initializer]
[--internal_noise INTERNAL_NOISE] [--trial TRIAL]
[--label LABEL] [--seed SEED] [-t] [--recalculate_statistics]
[--epochs EPOCHS] [--batch BATCH] [--image_path IMAGE_PATH]
[--train_image_path TRAIN_IMAGE_PATH] [--test_generalisation]
[--invert_test_images INVERT_TEST_IMAGES]
[--test_perturbations] [--data_augmentation]
[--extra_augmentation] [-c] [--skip_test] [-l] [--save_images]
[-p] [--gpu GPU] [--project_dir PROJECT_DIR] [-v VERBOSE]
optional arguments:
-h, --help show this help message and exit
--convolution CONVOLUTION
Name of convolutional filter to use
--base BASE Name of model to use
--pretrain Flag to use pretrained ImageNet weights in the model
--architecture ARCHITECTURE
Parameter file (JSON) to load
--interpolation INTERPOLATION
Method to interpolate the images when upscaling.
Default: 0 ("nearest" i.e. no interpolation)
--optimizer {SGD,RMSprop,Adagrad,Adadelta,Adam,Adamax,Nadam}
Name of optimizer to use: https://keras.io/optimizers/
--lr LR, --learning_rate LR
Learning rate for training
--decay DECAY Optimizer decay for training
--use_initializer Flag to use the weight initializer (then freeze
weights) for the Gabor filters
--internal_noise INTERNAL_NOISE
Standard deviation for adding a Gaussian noise layer
after the first convolutional layer
--trial TRIAL Trial number for labeling different runs of the same
model
--label LABEL For labeling different runs of the same model
--seed SEED Random seed to use
-t, --train Flag to train the model
--recalculate_statistics
Flag to recalculate normalisation statistics over the
training set
--epochs EPOCHS Number of epochs to train model
--batch BATCH Size of mini-batches passed to the network
--image_path IMAGE_PATH
Path to image files to load
--train_image_path TRAIN_IMAGE_PATH
Path to training image files to load
--test_generalisation
Flag to test the model on sets of untrained images
--invert_test_images INVERT_TEST_IMAGES
Flag to invert the luminance of the test images
--test_perturbations Flag to test the model on perturbed images
--data_augmentation Flag to train the model with data augmentation
--extra_augmentation Flag to train the model with additional data
augmentation
-c, --clean Flag to retrain model
--skip_test Flag to skip testing the model
-l, --log Flag to log training data
--save_images Flag to save preprocessed (perturbed) test images
-p, --save_predictions
Flag to save category predictions
--gpu GPU GPU ID to run on
--project_dir PROJECT_DIR
Path to the root project directory
-v VERBOSE, --verbose VERBOSE
Verbosity level

To train and test the models, the code below may be used and adapted as required.

importosimportsysimportpprintimportsubprocessimportrandomfromtqdm.notebookimporttqdmimporttensorflowastfimporttensorflow.kerasimporttensorflow.keras.backendproject_root_dir="/home/jovyan/work/BioNet"# Change as necessaryprint(f"Project directory: {project_root_dir}\n")
sys.path.append(project_root_dir)
print("\nTensorFlow:", tf.__version__)
print(f"Channel ordering: {tf.keras.backend.image_data_format()}") # TensorFlow: Channels last order.gpus=tf.config.experimental.list_physical_devices('GPU')
# gpus = tf.config.list_physical_devices('GPU')pprint.pprint(gpus)
label="paper"image_path=''# Empty string defaults to CIFAR-10# image_path = '/shared/data/ecoset-cifar10'convolutions= ['Original', 'Low-pass', 'DoG', 'Gabor', 'Combined-trim']
bases= ['ALL-CNN', 'VGG-16', 'VGG-19', 'ResNet']
seed=0start_trial=1num_trials=5trials=range(start_trial, start_trial+num_trials)
train=Truepretrain=Falseclean=Falseepochs=100optimizer="RMSprop"lr=1e-4use_initializer=Truedata_augmentation=Trueextra_augmentation=Falseinternal_noise=0skip_test=Falsesave_images=Falsesave_predictions=Truetest_generalisation=Truetest_perturbations=Trueinterpolation=4# Lanczosrecalculate_statistics=Falseverbose=0halt_on_error=Falsegpu=1######################################script=os.path.join(project_root_dir, "model.py")
flags= ['--log']
iftrain:
flags.append('-t')
ifclean:
flags.append('-c')
ifuse_initializer:
flags.append('--use_initializer')
ifdata_augmentation:
flags.append('--data_augmentation')
ifextra_augmentation:
flags.append('--extra_augmentation')
ifskip_test:
flags.append('--skip_test')
ifrecalculate_statistics:
flags.append('--recalculate_statistics')
ifsave_predictions:
flags.append('--save_predictions')
optional_args= []
ifimage_path:
optional_args.extend(['--image_path', str(image_path)])
iftest_perturbations:
optional_args.append('--test_perturbations')
iftest_generalisation:
optional_args.append('--test_generalisation')
ifpretrain:
optional_args.append('--pretrain')
ifinternal_noise:
optional_args.extend(['--internal_noise', str(internal_noise)])
ifinterpolation:
optional_args.extend(['--interpolation', str(interpolation)])
ifverbose:
optional_args.extend(['--verbose', str(verbose)])
count=1fortrialintqdm(trials, desc='Trial'):
ifseedisNone:
seed=random.randrange(2**32)
forbaseintqdm(bases, desc='Model Base', leave=False):
forconvintqdm(convolutions, desc='Convolution', leave=False):
cmd= [script, *flags]
ifsave_imagesandcount==1:
cmd.append('--save_images')
cmd.extend(['--convolution', conv, '--base', base, '--label', label,
'--trial', str(trial), '--seed', str(seed),
'--optimizer', optimizer, '--lr', str(lr),
'--epochs', str(epochs), '--gpu', str(gpu)])
cmd.extend(optional_args)
completed=subprocess.run(cmd, shell=False, capture_output=True, text=True)
ifcompleted.returncode!=0:
print(completed.stdout)
print(completed.stderr)
count+=1f'Finished job "{label}"!'

Notes

rsync -vzhrLKe ssh --progress user@host:/storage/models/paper /shared/data/

Create symlinks to consolidate simulations

find response -maxdepth 1 -mindepth 1 -type d -exec ln -s ../'{}' paper/ \;