Research map / Video / Action recognition
Diagnosing Error in Temporal Action Detectors
Humam Alwassel, Fabian Caba Heilbron, Victor Escorcia, Bernard Ghanem
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| limit predictions to top-10G instead of all (ablation) | SC · ActivityNet v1.3 | average-mAP_N (%): 33.92 -> 33.45 | No clear effect |
| limit predictions to top-10G instead of all (ablation) | BU · ActivityNet v1.3 | average-mAP_N (%): 17.26 -> 17.02 | No clear effect |
| use normalized mAP_N instead of mAP | SC · ActivityNet v1.3 | average mAP (%): 33.42 -> 33.92 | Improved |
| fix localization errors by removing those predictions (ablation) | all four detectors · ActivityNet v1.3 | average-mAP_N | Improved |
| fix other error types (double detection, wrong label, confusion, background) (ablation) | all four detectors · ActivityNet v1.3 | average-mAP_N | No clear effect |
| apply stricter NMS | SC · ActivityNet v1.3 | double detection error rate | Improved |
| restrict to instances with high agreement (H-XH) (ablation) | CES · ActivityNet v1.3 | average-mAP_N | Improved |
| restrict to instances with large context size (5-6) (ablation) | CES · ActivityNet v1.3 | average-mAP_N | Hurt |
| restrict to short instances (coverage or length) (ablation) | detectors · ActivityNet v1.3 | average-mAP_N | Hurt |
| add another instance of the same class in the video | detectors · ActivityNet v1.3 | miss detection rate | Hurt |
| vary annotator agreement (ablation) | detectors · ActivityNet v1.3 | average-mAP_N sensitivity | No clear effect |
Closest papers by meaning
- Temporal Action Localization in the Deep Learning Era: A Survey (TPAMI 2023 · similarity 0.92)
- Action Search: Spotting Actions in Videos and Its Application to Temporal Action Localization (ECCV 2018 · similarity 0.91)
- Graph Convolutional Module for Temporal Action Localization in Videos (TPAMI 2021 · similarity 0.91)
- Learning To Refactor Action and Co-Occurrence Features for Temporal Action Localization (CVPR 2022 · similarity 0.90)
- Benchmarking the Robustness of Temporal Action Detection Models Against Temporal Corruptions (CVPR 2024 · similarity 0.89)
- Test-Time Zero-Shot Temporal Action Localization (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
