Sift-flow evaluation bug for Fully Convolutional Neural Network

[Sharif Amit]

Hello,

I wanted to train my net on the sift-flow data. As done in FCN semantic segmenation.
But the author says "care must be taken for proper evalution by excluding the missing classes.".
So the training is smooth. And the evaluation result for Geometric labels is giving good outpus.
But there seems to be some bug for  Semantic Label evaluation.

And the terminal shows this runtime error

For mean accuracy :
/home/sharif_amit/Desktop/caffe/score.py:48: RuntimeWarning: invalid value encountered in true_divide
  acc = np.diag(hist) / hist.sum(1)

For mean IU :
/home/sharif_amit/Desktop/caffe/score.py:51: RuntimeWarning: invalid value encountered in true_divide
  iu = np.diag(hist) / (hist.sum(1) + hist.sum(0) - np.diag(hist))

The score layer that its indicating is this python file

Does anyone know how to fix this or any workaround for Sift-flow evaluation?

Thanks in advance