KimRass / DeepLabv3

PyTorch implementation of 'DeepLabv3' (Chen et al., 2018) and training it on VOC 2012

'Rethinking Atrous Convolution for Semantic Image Segmentation'

1. Pre-traind Models

deeplabv3-voc2012.pt	- Trained on VOC 2012 training set for 54,000 steps. - Average mIoU on VOC 2012 validation set: 0.6563

2. Prediction on VOC 2012 Validation Set

Ground truth	Prediction

Ground truth	Prediction