yayuanli/MATT-Bench

Mistake Attribution: Fine-Grained Mistake Understanding in Egocentric Videos

CVPR 2026

Yayuan Li¹, Aadit Jain¹, Filippos Bellos¹, Jason J. Corso^1,2

¹University of Michigan, ²Voxel51

[Paper] [Code] [Project Page]

---

Dataset coming soon. We are preparing the data for public release. Stay tuned!

MATT-Bench Overview

MATT-Bench provides two large-scale benchmarks for Mistake Attribution (MATT) — a task that goes beyond binary mistake detection to attribute what semantic role was violated, when the mistake became irreversible (Point-of-No-Return), and where the mistake occurred in the frame.

The benchmarks are constructed by MisEngine, a data engine that automatically creates mistake samples with attribution-rich annotations from existing egocentric action datasets:

Dataset	Samples	Instruction Texts	Semantic	Temporal	Spatial
Ego4D-M	257,584	16,099	✓	✓	✓
EPIC-KITCHENS-M	221,094	12,283	✓	—	—

These are at least two orders of magnitude larger than any existing mistake dataset.

Annotations

Each sample consists of an instruction text and an attempt video, annotated with:

• Semantic Attribution: Which semantic role (predicate, object) in the instruction is violated in the attempt video
• Temporal Attribution: The Point-of-No-Return (PNR) frame where the mistake becomes irreversible (Ego4D-M)
• Spatial Attribution: Bounding box localizing the mistake region in the PNR frame (Ego4D-M)

Citation

@inproceedings{li2026mistakeattribution,
  title     = {Mistake Attribution: Fine-Grained Mistake Understanding in Egocentric Videos},
  author    = {Li, Yayuan and Jain, Aadit and Bellos, Filippos and Corso, Jason J.},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2026},
}