AI Paper Atlas

Research map / Video / Video segmentation and tracking

Fine-VAD: Towards Fine-Grained Video Anomaly Detection via Progressive Cross-Granularity Learning

Menghao Zhang, Yiyan Zhu, Pengfei Ren, Haifeng Sun, Qi Qi, Zirui Zhuang, Huazheng Wang, Lei Zhang, Jianxin Liao, Jingyu Wang

CVPR 2026 · Video › Video segmentation and tracking · approach: Weakly supervised video anomaly detection

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 changedWhereMeasuredEffect
add coarse-level binary alignment to fine-only training (ablation)Fine-VAD · UCF-CrimeAVG mAP (%): 8.64 -> 10.93Improved
add intermediate-level pseudo macro-category guidance to fine-only training (ablation)Fine-VAD · UCF-CrimeAVG mAP (%): 8.64 -> 11.60Improved
use all three levels (coarse, intermediate, fine) together (ablation)Fine-VAD · UCF-CrimeAVG mAP (%): 10.93 -> 14.99Improved
remove the temporal adapter (ablation)Fine-VAD · UCF-CrimeAVG mAP (%): 14.99 -> 5.62Hurt
swap LGT-Adapter for Transformer, GCN, Axial TF or TA adapters (ablation)Fine-VAD · UCF-CrimeAVG mAP (%): 14.99 -> 12.01 to 14.56Hurt
add coarse-level contrastive loss Lcts (ablation)Fine-VAD · UCF-CrimeAVG mAP (%): 10.93 -> 12.90Improved
add supervised loss (Lmce or Lmce+Lcts) at intermediate level (ablation)Fine-VAD · UCF-CrimeAVG mAP (%): 14.99 -> 12.58 / 11.96Hurt
add contrastive loss at fine level on top of Lrefine (ablation)Fine-VAD · UCF-CrimeAVG mAP (%): 14.99 -> 14.86No clear effect
add contrastive loss Lcts at coarse level together with Lrefine (ablation)Fine-VAD · UCF-CrimeAVG mAP (%): 12.07 -> 14.99Improved
vary number of pseudo macro-categories K (2,3,4,5) (ablation)Fine-VAD · UCF-Crime and XD-ViolenceAVG mAP (%): K=4: 14.99 / 31.87; K=2: 9.14 / 27.74; K=5: 12.37 / 25.88Mixed
apply progressive cross-granularity paradigm to I3D (FineCNN)I3D baseline · UCF-CrimeAVG mAP (%): 3.24 -> 6.21Improved
apply progressive cross-granularity paradigm to I3D (FineCNN)I3D baseline · XD-ViolenceAVG mAP (%): 11.65 -> 13.58Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy