/semantic-segmentation-transfer-learning

Semantically segmenting images using Fully Convolutional Networks based on VGG-16 network in PyTorch

Primary LanguageJupyter Notebook

Semantic Segmentation using Fully Convolutional Networks (FCN - 32/16)

  • I have implemented two FCNs (FCN-32 and FCN-16) to semantically segment images using the VOCSegmentation2012 dataset.
  • Semantic segmentation is a bit different from classification, where we classify each pixel as a particular class.
  • I've used Transfer Learning by transfering initial weigths of VGG-16 network and replacing classification layer with convolution layers.
  • mean IOU and DICE score are the evaluation metrics.