bharat-b7/unsupervised-learning-of-deep-feature-representation-for-clustering-egocentric-actions

unsupervised-learning-of-deep-feature-representation-for-clustering-egocentric-actions's Issues

How did you get the Figure1

Hi.

I have a question about your paper.
You showed splices make classes in high dimensional feature space in Figure1.
I could not reproduced Figure1 by t-SNE. How did you get Figure1?

the number of classes of ADL-short dataset is not 21 but 20

In your paper, you wrote the number of classes of ADL-short dataset is 21.
But actually, I counted the annotation (http://cvit.iiit.ac.in/projects/FirstPersonActions/)
I could find only 20 classes.

"take"  "wash"  "pour"   "open"   "put"  "brush"  "close"  "wipe"  "stir"  "type" 
"dip"  "write"  "wear" "spray" "flip" "swtichon" "tear" "comb" "scoop" "bg" 

About annotation of ADL-long dataset.

Thank you for your providing source code of the paper.
It helps us a lot!!

I have a question about your experiment.
According to your paper, you evaluated your method on ADL-long and you wrote the class number of ADL-long is 12. I downloaded ADL dataset from this page and I found the number of classes are 32.

I want to see ground truth labels of ADL-long dataset annotated 21 class. Where can I see it?

Best,