IoU should ignore void class

Hello, I think the provided code in evaluate.py does not ignore the void label in ground truth masks (which is mapped to the background in the 1D labels); therefore, the performance is underestimated compared to the results of VOC devkit. Here is the link to the line ignoring void pixels in the VOC devkit.

[DrSleep]

I can confirm that with weights set the results are improved: from 76.8 mIoU to 80.1 mIoU on the validation set. I have used the code below, and will update the evaluation script accordingly. Thank you for letting know about the issue!

pred = tf.reshape(pred, [-1,])
gt = tf.reshape(label_batch, [-1,])
weights = tf.cast(tf.less_equal(gt, 20), tf.int32)

# mIoU
mIoU, update_op = tf.contrib.metrics.streaming_mean_iou(pred, gt, num_classes=21, weights=weights)

[DrSleep]

I have pushed the changes to the master branch. Thanks again!