Challenge Details

Data & Protocol

Numerous studies have reported high recognition rates of up to 90% for the classification of high pain intensity levels. However, the reliable detection and differentiation of low and medium pain levels remains a persistent challenge, as physiological, facial, or vocal indicators are often more subtle, ambiguous, and subject to strong inter-individual variation. The "X-ITE Pain" challenge 2025 therefore places a particular emphasis on the accurate classification of low and moderate pain intensities - a crucial step toward building robust and clinically meaningful pain assessment systems.

The "X-ITE Pain" challenge 2025 data set contains a total of 30 subjects. The data set is divided into two groups for the challenge: Training (26 subjects) and Testing (4 subjects). The objective is to classify the unseen data of the 4 test subjects based on short multimodal recordings (segments) into one of two categories: "low pain" (PL_1) and "medium pain" (PL_2).

Labeled training set is now accessible, while the new, unlabeled test set will be provided on June 20, 2025.

Training Set (Labeled) :
Participants will receive data from 26 individuals, each recorded under two distinct pain conditions:
- PL_1 – low pain
- PL_2 – medium pain

For each subject and condition, 60 segments of 10 seconds are provided, resulting in:
- 120 labeled segments per subject
- 3,120 labeled segments in total

All segments include synchronized biosignal, video, and audio data. Ground-truth labels for pain intensity are included.
(Note: Bio data will be provided as .mat-files, videos as .mp4, audio as .mp3)

Test Set (Unlabeled) :
The core challenge is to generalize to new individuals. Participants will be provided with data from 4 additional subjects under the same experimental conditions:
- 2 pain levels × 60 segments = 120 unlabeled segments per subject
- 480 test segments in total

The structure, modalities, and format are identical to the training set, but ground-truth labels are withheld. Teams must predict the "correct" pain intensity (PL_1, PL_2) for each individual segment in the test set.

Modalities / Signals :

Example of a 10s segment (for illustration purposes only, pain intensity unknown)

➜ First results with the "X-ITE Pain" db can also be found in:
doi.org/10.1109/ACIIW.2019.8925061

There are no separate challenges related to biosignals only, audio only or video only. Participants have the freedom to use either one modality (signal), a range of modalities or all modalities. The teams are encouraged to employ their very own features and classification techniques. Labels of the test set will be undisclosed until the end of the challenge. While papers may include results obtained from the training set, only outcomes from the test set will contribute to the overall "X-ITE Pain" challenge results ranking.

All participants in the "X-ITE Pain" challenge are expected to submit class-specific probability scores for each test segment in addition to their final label predictions to the organizers by June 30, 2025. (These per-segment probabilities offer valuable insight into model confidence and decision boundaries, and may be used for secondary analyses such as calibration assessment, ranking, or tie-breaking.) Final results will be presented at the Challenge Workshop (October 11, 2025), held in conjunction with the ACII 2025 in Canberra, Australia. In addition, each team must submit a paper (short papers are welcome but not obligatory) outlining their methodology, findings, and insights gained through the challenge, which will also be featured at the workshop.
For further information, please visit the Submission page.

To register your team, go to the Challenge Registration.

Page updated

Google Sites

Report abuse