Question about the labels_edin_locomotion.zip file and ELMD data

[whjo97]

@UttaranB127

Dear Uttaran.
I analyzed the labels_edin_locomotion.zip file provided with the latest dataset, but the labels don't match.

I merged the csv files of labels_edin_locomotion.zip to obtain the number of class votes per sample.

I then merged them with the labels_ELMD.h5 file to obtain the number of class votes per label.

In the closed issue #11, you mentioned that the labels are mapped as [0, 1, 2, 3] -> ['Angry', 'Neutral', 'Happy', 'Sad'].
However, comparing the labels_edin_locomotion file with the labels_ELMD file, it appears to be [0, 1, 2, 3] -> ['Happy', 'Sad', 'Angry', 'Neutral'].
As I understand it, the labels and labels_ELMD files are where the authors labeled perceived emotions, while the labels_edin_locomotion file appears to indicate the classes voted on by the other 10 annotators.
However, even if the annotators were different, there are significant discrepancies. Why is this discrepancy occurring?

Also, the ELMD website states that there are 21 joints and that they were collected at 60Hz, but the provided data only has 16 joints.
As for the number of joints, was the data preprocessed differently to have 16 joints?
Were the 342 samples collected by the authors also collected at 60Hz?
The paper says they walked 7 meters, but the sequence is too short.

Thank you.

[UttaranB127]

Hi,

Please feel free to reuse the dataset and scripts as you see fit, since the original dataset and scripts are not actively maintained since the last 6+ years. To summarize your queries:

• the dataset is a combination of multiple smaller datasets, and the labels were merged to create the final 4 labels. The script for merging may be outdated or unavailable.

• the joint motions were collected on video at 60 Hz, but the dataset contains filtered and cleaned joint positions at a much lower frame, probably somewhere between 15 and 20 Hz. This is one reason for seeing shorter sequences than what would be expected at 60 Hz.

• 21-joint motions were downsampled to 16 by merging certain compatible joints. The corresponding scripts may also be outdated or unavailable.

[whjo97]

Thank you for your kind response!