CODEWITHSUNDEEP
tensorflow
Ray.R.Chua
Ray.R.Chua

Reputation: 777

Can anyone give an example on how tf.contrib.metrics.streaming_mean_iou in tensorflow works?

I am using tensorflow 0.11 and am running some image segmentation tests. In image segmentation, we usually compute the IoU. How can I make use of tensorflow's tf.contrib.metrics.streaming_mean_iou?

mIoU, confusionMatrix = tf.contrib.metrics.streaming_mean_iou(pred_labels_vec,gt_labels_vec,NUM_CLASS)

init = tf.initialize_local_variables()

sess.run(init)

for i in range(1000):
  iou, _ = sess.run([mIoU, confusionMatrix])
  print(iou)

Upvotes: 1

Views: 1543

Answers (1)

LI Xuhong
LI Xuhong

Reputation: 2356

Your code is correct to compute the mean iou.

If you want to compute the IoU for each class, see this question. You can compute them via the confusion matrix. When you run streaming_mean_iou, tensorflow has already created a local variable, called total_confusion_matrix (same for tf 0.11). Then you can get this matrix by calling tf.local_variables(), or through get_tensor_by_name.

Upvotes: 2

Related Questions