what is input data for training detector?

[learningsteady0J0]

Hi!

I read your paper well. I will use this to recognize Korean sign language.

I don't know if I didn't find it, What input data did you put into the detector?

I think that EX)

gesture : Use all 8 frames starting within the dataset.

no_gesture : i don't know...

Thank you so much for your good experiment!

[ahmetgunduz]

Hi @learningsteady0J0 , the data is frames of videos. You can use different modalities (Depth, RGB or RGB-D) the same way in the classifier. The only difference between classifier and dedector is the number of classes. The detector is a binary classifier.

[learningsteady0J0]

between

Thank you for your reply! I don't think I got it right.

If class is up to 1,2,3...9 in the dataset A, 'classifier' use class 1,2,3...9 but 'detector' use class gesture(all 1 to 9) and no gesture.

Is that what I thought? So how did you collect the dataset for 'no gesture'?

I don't know if I wrote it well because I'm not good at English. Thank you so much!

[ahmetgunduz]

If class is up to 1,2,3...9 in the dataset A, 'classifier' use class 1,2,3...9 but 'detector' use class gesture(all 1 to 9) and no gesture That is right!

Egogesture, NVGesture and Jester datasets has either has "no gesture" class or I extracted the no gesture parts. If you do notha have "no gesture" classes in your dataset, you can train a detector using a similar dataset with no gesture classes.

[learningsteady0J0]

You made everything clear.

Thank you veryyyyyyy much!

I hope everything you do goes well!