Doppio: TU Darmstadt dataset trains vision models to weigh falling coffee grounds contactless
rsasaki0109 · x · 2026-09-07
Researchers from TU Darmstadt, TUM and hessian.AI present Doppio, a GCPR 2026 dataset and study on contactless mass estimation via computer vision.
- Motivation: Industrial powder weighing usually needs scales, but many applications require contactless sensing where existing solutions are costly and application-specific.
- Dataset: Using coffee grinding as a case study, videos of falling grounds are paired with per-frame weight annotations extracted by OCR from the scale display, smoothed and time-lag compensated.
- Experiments: Deep learning baselines from purely spatial feed-forward networks to recurrent spatio-temporal models are evaluated for accuracy and compute trade-offs.
- Findings: Vision models can accurately estimate the cumulative weight of falling particles, establishing a foundation for contactless measurement.
Paper, dataset and code are publicly available.
More from Research
- Principia benchmark: no video model exceeds 0.42 on physics despite VBench ~0.8 — anand_bhattad · 2026-09-07
- HiSfM: scaffold-anchored hierarchical SfM tames repeated-structure ambiguity and cuts runtime — zhenjun_zhao · 2026-09-07
- BLASt3R (ECCV'26): uncalibrated bundle adjustment beats all prior calibrated VSLAM methods — zhenjun_zhao · 2026-09-07
- Economists keep getting overparameterization wrong: SGD's implicit regularization is the point — Afinetheorem · 2026-09-07
- Diophantine imaging paper reveals broken symmetry in sums of three cubes — CatAstro_Piyush · 2026-09-07
- Empirical check finds AI detectors do not disproportionately flag autistic people's writing as AI — BlancheMinerva · 2026-09-07