N950 / edge-segnet

EdgeSegNet - Pytorch

An implementation of EdgeSegNet in Pytorch, CamSeg01 dataset is used for training

Repo containes :

• EdgeSegNet.ipynb Notebook taking care of downloading the dataset, training, plotting and prediction examples.

• train.py entry point script which does all the downloading and setting up of the dataset, training and plotting graphs and prediction example

• EdgeSegNet.py is currently implemented in the exact architecture detailed in the paper, but could be modified easily.

• NetworkModules.py custom modules are also implemented as Pytorch modules.

• CamSeqDataset.py a Pytorch Dataset for CamSeg01, downloads and unzips imgs.

• dataset_backup

train.py

Default params will yield 90% val accuracy withing 40-50 epochs, 2 min on colab gpu

usage: train.py [-h] [--learning-rate lr] [--batch-size B] [--n_epochs N]
                [--gamma G] [--scheduler-step S]

A Training script for EdgeSegNet :: https://arxiv.org/abs/1905.04222

optional arguments:
  -h, --help          show this help message and exit
  --learning-rate lr  Default 0.001, initial learning rate for the Adam optimizer, scheduled by StepLR
  --batch-size B      Default 16, Batch size for both train and validation, keep in mind the dataset has a total of 101 imgs only
  --n_epochs N        Default 50, number of training epochs
  --gamma G           Default 0.95, Multiplicative factor of learning rate decay
  --scheduler-step S  Default 25, Scheduler step, each S epochs learning rate is updated