ESTANet: Efficient Online Error Detection in Procedural Videos via Prediction Inconsistency
Shih-Po Lee, Reza Ghoddoosian, Faizan Siddiqui, Enna Sachdeva, Behzad Dariush · 2026
Read ESTANet: Efficient Online Error Detection in Procedural Videos via Prediction Inconsistency by Shih-Po Lee, Reza Ghoddoosian, Faizan Siddiqui, Enna Sachdeva, Behzad Dariush (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-25317.txt.
Description
An assistant watching someone follow a procedure should flag mistakes the moment they happen, using only frames seen so far. Shih-Po Lee and colleagues propose ESTANet in 2026, built on a simple observation about action detectors. Detectors tuned differently, some standard and some deliberately error sensitive, agree when a step is performed correctly. When an error occurs, their predictions diverge, and using varied time windows widens the disagreement further. Majority voting over the mismatches marks the faulty frames.
Tested on three public datasets of assembly and everyday tasks, the framework reached state of the art accuracy for online error detection. It stayed lightweight enough for real time use, avoiding the heavy custom architectures earlier systems relied on. The authors present prediction inconsistency itself as a practical error signal that needs no specialised supervision to train.
Catalog record
| CREATOR | Shih-Po Lee, Reza Ghoddoosian, Faizan Siddiqui, Enna Sachdeva, Behzad Dariush |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.25317