audiovisualemotionchallenge / avec2019 Goto Github PK

Excuse me, I have searched throgh the AVEC2019 workshop and baseline paper for a long time, still I didn't get the way to obtain the data set for DDS.

I am trying to reproduce the results on your baseline paper of subtask CES.
I only use audio_egemaps_xbow feature extracted by the scripts you provided, and finish the experiment using the code of provided baseline system . But I got lower ccc score than the paper reported.
The followings are what I got, and all the ccc score is evaluated on development set.
Arousal:
DE+HU: 0.342 DE: 0.374 HU:0.231
Valence:
DE+HU: 0.274 DE: 0.373 HU: 0.058
Liking:
DE+HU: 0.191 DE: 0.151 HU: 0.232

And the following are the ccc scores reported on your baseline paper of BoAW-e(TABLE 7)
Arousal:
DE+HU: 0.398 DE: 0.434 HU: 0.291
Valence:
DE+HU: 0.352 DE: 0.455 HU: 0.135
Liking:
DE+HU: 0.138 DE: 0.003 HU: 0.253

Any suggestions about it?

What is the expected structure of the data directory for DDS? It looks like the DAIC-WOZ dataset comes with a single folder for each participant... is this is what is required for feature extraction? Thank you!

Hello, I encountered a problem while reproducing the cross-cultural sub project (CES) of this project. I was unable to find the accurate training set, testing set, and validation set. The description of the dataset in Table 3 of the paper is video chat, so the activity chose video chat;
Data training and validation sets require labels such as Valence, Arousal, and Liking. Three datasets are available for all three labels: 30 in Germany and 35 in Hungary, but in the paper, there are only 34 training sets, plus 14 validation sets.
Can you tell me how to select data and partition them? It would be even better if a dataset could be provided. My email is [email protected] . Looking forward to your reply.

Hi, I am trying to access the CES partition for 2019 challenge and have also tried signing up for the data download. Could you please let me know how I can access the partition along with the combined annotations (gold standard) from different raters for the CES challenge?

Based on the default setting, my ccc score computed by resnet deep learning features is 0.05292, which is much lower than the baseline 0.269