Multimodal Emotion Recognition Challenge by Neurodata Lab
Hello! Our laboratory Neurodata Lab launches in October-November 2017 its own ML-contest for recognizing emotions.
→ Competitive area
The contest itself, called Multimodal Emotion Recognition Challenge (MERC 2017) , will be held in the online format.
At this stage, the goal is to recognize in a video the basic emotions of a person through four channels — face, voice, eyes, body — by which the required signs are calculated (which we will tell more about by the start of the competition). The main emphasis is placed on the fact that some data remain inaccessible, which means you need to learn how to make a correct prediction in such conditions (about 30% of the data in the sample is missing - they are replaced by zeros).
')
Output: 6 emotions. Each person is marked up by annotators according to what emotion he is experiencing at a given time. At the training sample, the agreement score will also be available, calculated as the ratio of “for” tags the most frequent emotion (it is always the same) to the total number of tags placed by several annotators.
We have a leaderboard. It is formed according to the metric unweighted accuracy. In more detail - again after the official start. In addition to the general prize fund ($ 3,500.00), the jury members have the right to note the decisions they like most, based on three criteria: accuracy, versatility and innovation.
This is the first of a series of our initiatives (competitive and grant), called upon gradually and with the complication from step to step to introduce a wide range of Emotion AI (emotional AI, as the direction of "emotional research and technology" is increasingly called) in the current context of reflection, analysis and the development of the Russian-speaking (and not only) community of specialists in Data Science, ML and CV.
Pre-registration is open on October 6th.
Access to the materials will be provided on October 18, 2017.
Summing up and announcement of the winners are expected on November 28, 2017.
The competition jury includes experts from Skoltech-IITP RAS and NRU ITMO; the project is provided with information and consulting support by colleagues from MIPT, VMK MGU and Politecnico di Milano University of Milan.
Join now! And success!
→ Competitive area
The contest itself, called Multimodal Emotion Recognition Challenge (MERC 2017) , will be held in the online format.
At this stage, the goal is to recognize in a video the basic emotions of a person through four channels — face, voice, eyes, body — by which the required signs are calculated (which we will tell more about by the start of the competition). The main emphasis is placed on the fact that some data remain inaccessible, which means you need to learn how to make a correct prediction in such conditions (about 30% of the data in the sample is missing - they are replaced by zeros).
')
Output: 6 emotions. Each person is marked up by annotators according to what emotion he is experiencing at a given time. At the training sample, the agreement score will also be available, calculated as the ratio of “for” tags the most frequent emotion (it is always the same) to the total number of tags placed by several annotators.
We have a leaderboard. It is formed according to the metric unweighted accuracy. In more detail - again after the official start. In addition to the general prize fund ($ 3,500.00), the jury members have the right to note the decisions they like most, based on three criteria: accuracy, versatility and innovation.
This is the first of a series of our initiatives (competitive and grant), called upon gradually and with the complication from step to step to introduce a wide range of Emotion AI (emotional AI, as the direction of "emotional research and technology" is increasingly called) in the current context of reflection, analysis and the development of the Russian-speaking (and not only) community of specialists in Data Science, ML and CV.
Pre-registration is open on October 6th.
Access to the materials will be provided on October 18, 2017.
Summing up and announcement of the winners are expected on November 28, 2017.
The competition jury includes experts from Skoltech-IITP RAS and NRU ITMO; the project is provided with information and consulting support by colleagues from MIPT, VMK MGU and Politecnico di Milano University of Milan.
Join now! And success!
Source: https://habr.com/ru/post/339532/
All Articles