Research map / Vision-language and multimodal / Visual question answering
Free VQA Models from Knowledge Inertia by Pairwise Inconformity Learning
Yiyi Zhou, Rongrong Ji, Jinsong Su, Xiangming Li, Xiaoshuai Sun
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 answer embedding loss only (ablation) | Baseline · VQA2.0 val | accuracy (All): 60.37 -> 60.74 | Improved |
| add full Pairwise Inconformity Learning (ablation) | Baseline · VQA2.0 val | accuracy (All): 60.37 -> 61.56 | Improved |
| add hard example mining to PIL (ablation) | Baseline+PIL · VQA2.0 val | accuracy (All): 61.56 -> 62.01 | Improved |
| add PIL with Faster R-CNN features (ablation) | Baseline+FRCNN · VQA2.0 val | accuracy (All): 63.72 -> 64.58 | Improved |
| add hard example mining with FRCNN features (ablation) | Baseline+PIL+FRCNN · VQA2.0 val | accuracy (All): 64.58 -> 64.63 | No clear effect |
| increase beta from 0.25 to 1 (alpha=0.01) (ablation) | Baseline+PIL · VQA2.0 val | accuracy (All): 61.56 -> 61.06 | Hurt |
| decrease beta to 0.05 (alpha=0.01) (ablation) | Baseline+PIL · VQA2.0 val | accuracy (All): 61.56 -> 60.84 | Hurt |
| set alpha=0.25 (beta=0.25) (ablation) | Baseline+PIL · VQA2.0 val | accuracy (All): 61.56 -> 61.50 | No clear effect |
| l2 regularization on joint features instead of PIL (ablation) | Baseline · VQA2.0 val | accuracy (All): 60.37 -> 59.47 | Hurt |
| contrastive loss on joint features (ablation) | Baseline · VQA2.0 val | accuracy (All): 60.37 -> 59.58 | Hurt |
| KL-divergence regularization on predictions (ablation) | Baseline · VQA2.0 val | accuracy (All): 60.37 -> 59.18 | Hurt |
| apply PIL on attention feature fv instead of joint feature (ablation) | Baseline+PIL · VQA2.0 val | accuracy (All): 61.56 -> 61.32 | Hurt |
Closest papers by meaning
- Inverse Visual Question Answering: A New Benchmark and VQA Diagnosis Tool (TPAMI 2018 · similarity 0.92)
- Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · similarity 0.92)
- Tips and Tricks for Visual Question Answering: Learnings From the 2017 Challenge (CVPR 2018 · similarity 0.91)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.91)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.91)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.90)
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.
