Result on testImage.jpg is wrong

[riverfan]

Dear deepmask inventors,

I have compiled torch through its install.sh script after a fresh install of cuda 8.0.61 and cudnn 5.1 . The torch compilation worked without errors, and I have installed the coco API and inn modules.

I have been able to execute the command lines from the README to download the models and execute a segmentation on the 'data/testImage.jpg' example. The resulting image was not what I expected:

Any idea on where to start in order to understand what went wrong? Did anyone have a similar experience?

Thanks you in advance for your answer.