The output from demo trained model seem uncorrect

[B-Paweekorn]

I'm attempting to run the demo in a Docker environment with the following specifications:
Ubuntu 16.04, Python 3.5, CUDA 8.0, and PyTorch 0.4.0. However, it appears that the segmentation isn't functioning as expected.