Multi-class semantic segmentation via Transfer Learning

This repository contains code for Fine Tuning DeepLabV3 ResNet101 in PyTorch. The model is from the torchvision module.

Dataset:

I've fine tuned the model for the eTRIMS Image Database data-set. I have used the eTRIMS Database beta 2, with 8 classes of objects(Building, Car, Door, Pavement, Road, Sky, Vegetation, Window)

Implementation Details:

The model was fine tuned for 40 epochs and achieves an testing loss value of 0.859484. The code was written using pyTorch and trained on NVIDIA Tesla k80 GPU with 12GB memory via the Google colab

Output:

The segmentation output of the model on a sample image are shown below.

Figure:(From top to bottom, left to right) Image, Building, Car, Door, Pavement, Road, Sky, Vegetation, window, void.

Loss plot is shown below.

Running the Code:

To run the code the dataset use the following command.

python main.py "data_directory_path" "experiment_folder_where weights and log file need to be saved"

It has following two optional arguments:

--epochs : Specify the number of epochs. Default is 25.
--batchsize: Specify the batch size. Default is 4.

Citation

eTRIMS Image Database

@techreport{ korc-forstner-tr09-etrims,
             author = "Kor{\v c}, F. and F{\" o}rstner, W.",
             title = "{eTRIMS} {I}mage {D}atabase for Interpreting Images of Man-Made Scenes",
             number = "TR-IGG-P-2009-01",
             month = "April",
             year = "2009",
             institute = "Dept. of Photogrammetry, University of Bonn",
             url = "http://www.ipb.uni-bonn.de/projects/etrims_db/" }

DeepLabv3 semantic segmentation architecture

@article{chen2017rethinking,
  title={Rethinking atrous convolution for semantic image segmentation},
  author={Chen, Liang-Chieh and Papandreou, George and Schroff, Florian and Adam, Hartwig},
  journal={arXiv preprint arXiv:1706.05587},
  year={2017}
}