Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories

TL;DR

This paper introduces a model-agnostic method for detecting annotation errors in videos by analyzing loss trajectories across training checkpoints, effectively identifying mislabeling and ordering issues without ground truth.

Contribution

The authors propose a novel approach using cumulative sample loss trajectories to detect annotation errors in video datasets, applicable across different datasets and models.

Findings

Effective detection of mislabeling and disordering in videos

Strong performance demonstrated on EgoPER and Cholec80 datasets

No ground truth required for error detection

Abstract

High-quality video datasets are foundational for training robust models in tasks like action recognition, phase detection, and event segmentation. However, many real-world video datasets suffer from annotation errors such as *mislabeling*, where segments are assigned incorrect class labels, and *disordering*, where the temporal sequence does not follow the correct progression. These errors are particularly harmful in phase-annotated tasks, where temporal consistency is critical. We propose a novel, model-agnostic method for detecting annotation errors by analyzing the Cumulative Sample Loss (CSL)--defined as the average loss a frame incurs when passing through model checkpoints saved across training epochs. This per-frame loss trajectory acts as a dynamic fingerprint of frame-level learnability. Mislabeled or disordered frames tend to show consistently high or irregular loss patterns, as they remain difficult for the model to learn throughout training, while correctly labeled frames typically converge to low loss early. To compute CSL, we train a video segmentation model and store its weights at each epoch. These checkpoints are then used to evaluate the loss of each frame in a test video. Frames with persistently high CSL are flagged as likely candidates for annotation errors, including mislabeling or temporal misalignment. Our method does not require ground truth on annotation errors and is generalizable across datasets. Experiments on EgoPER and Cholec80 demonstrate strong detection performance, effectively identifying subtle inconsistencies such as mislabeling and frame disordering. The proposed approach provides a powerful tool for dataset auditing and improving training reliability in video-based machine learning.