Research map / Vision-language and multimodal / Visual question answering
Overcoming Dual Drift for Continual Long-Tailed Visual Question Answering
Feifei Zhang, Zhihao Wang, Xi Zhang, Changsheng Xu
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 |
|---|---|---|---|
| add Balanced Classifier Prototype (ETF classifier) to the replay baseline (ablation) | replay baseline (ER-style) · CLT-VQA VQA v2 (ordered) | AP: 34.30 -> 39.82 | Improved |
| add Multi-modal Feature Alignment (OT) alone to the baseline (ablation) | replay baseline · CLT-VQA VQA v2 (ordered) | AP: 34.30 -> 35.79 | Improved |
| add MFA on top of BCP (ablation) | BCP model · CLT-VQA VQA v2 (ordered) | AP: 39.82 -> 40.69 | Improved |
| initialize classifier with ETF instead of nearest class mean (ablation) | BCP · CLT-VQA VQA v2 (ordered) | AP: 36.44 -> 39.82 | Improved |
| use optimal transport distance instead of cosine in MFA (ablation) | BCP & MFA · CLT-VQA VQA v2 (ordered) | AP: 40.52 -> 40.69 | Improved |
| use cosine distance instead of L2 in MFA (ablation) | BCP & MFA · CLT-VQA VQA v2 (ordered) | AP: 40.34 -> 40.52 | Improved |
| use long-tailed methods (LDAM, GCL) without anti-forgetting mechanism vs Vanilla | Vanilla CLT-VQA baseline · CLT-VQA VQA v2 (ordered) | AP: 15.83 -> 16.54 | Improved |
| use regularization-based methods (EWC, MAS) instead of replay-based methods | CLT-VQA · CLT-VQA VQA v2 (ordered) | AP: 24.12 -> 34.30 | Hurt |
| replace VQACL with the proposed method | VQACL · CLT-VQA VQA v2 (random) | AP: 37.89 -> 40.86 | Improved |
Closest papers by meaning
- MacVQA: Adaptive Memory Allocation and Global Noise Filtering for Continual Visual Question Answering (AAAI 2026 · similarity 0.93)
- VQACL: A Novel Visual Question Answering Continual Learning Setting (CVPR 2023 · similarity 0.93)
- Re-evaluating Continual VQA: Toward Fair and Robust Evaluation for Multimodal Continual Learning (CVPR 2026 · similarity 0.92)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.90)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.90)
- CL-MoE: Enhancing Multimodal Large Language Model with Dual Momentum Mixture-of-Experts for Continual Visual Question Answering (CVPR 2025 · similarity 0.89)
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.
