Introduction

End-to-end pipeline data augmentation and training and evaluation script using PyTorch

Data Augmentation

Generates different image transformation with many configurable parameters for scales, rotations, flips, crops and distortions with gaussian blurring, brightness and contrast variations.

Automated Training

Automated flexible model selection and evaluation with different learning rates, learning rate decay, checkpoints and train test split.

Usage

$ python striping.py


$ python train.py -h

usage: train.py [-h]
                [--model {alexnet,lenet5,stn-alexnet,stn-lenet5,capsnet,convneta,convnetb,convnetc,convnetd,convnete,convnetf,convnetg,convneth,convneti,convnetj,convnetk,convnetl,convnetm,convnetn,resnet18}]
                [--dataset {custom,cifar,hdf5}] [--num_classes NUM_CLASSES]
                [--batch_size BATCH_SIZE] [--transform TRANSFORM]
                [--num_workers NUM_WORKERS] [--lr-decay LR_DECAY]
                [--l2-reg L2_REG] [--hdf5-path HDF5_PATH]
                [--trainset-dir TRAINSET_DIR] [--testset-dir TESTSET_DIR]
                [--grey GREY]

PyTorch Automated Model Training & Evaluation

optional arguments:
  -h, --help            show this help message and exit
  --model {alexnet,lenet5,stn-alexnet,stn-lenet5,capsnet,convneta,convnetb,convnetc,convnetd,convnete,convnetf,convnetg,convneth,convneti,convnetj,convnetk,convnetl,convnetm,convnetn,resnet18}
  --dataset {custom,cifar,hdf5}
  --num_classes NUM_CLASSES
  --batch_size BATCH_SIZE
  --transform TRANSFORM
  --num_workers NUM_WORKERS
  --lr-decay LR_DECAY
  --l2-reg L2_REG
  --hdf5-path HDF5_PATH
  --trainset-dir TRAINSET_DIR
  --testset-dir TESTSET_DIR
  --grey GREY