Open science

Datasets & Benchmarks

LIVIA releases large-scale, carefully annotated datasets to push computer vision and affective computing research forward. Each comes with code, pretrained weights and reproducible benchmarks.

ICLR 2026 · ABAW @ ECCV 2026

BAH: Behavioural Ambivalence / Hesitancy in Videos

The first multimodal, subject-based video dataset for recognizing ambivalence and hesitancy (A/H), built for digital behavioural-change interventions.

300 participants · 1,427 videos. Frame- & video-level A/H annotations, multimodal (video, audio, text), aligned faces, transcripts and rich demographics. Now powering the 3rd AH Recognition Challenge at ABAW 11th (ECCV 2026).

Learn more GitHub Paper

NeurIPS 2024

SR-CACO-2: Confocal Microscopy Super-Resolution

A large-scale dataset for single-image super-resolution (SISR) in confocal fluorescence microscopy, with real low-resolution images captured directly by the microscope.

2,200 images · 9,937 patches. Real LR/HR pairs, three upscaling levels (×2, ×4, ×8), three fluorescent markers, ~16,800 multi-cellular objects and 16 benchmarked SISR methods.

Learn more GitHub Paper