Research map / Vision-language and multimodal / Visual question answering
Counterfactual Samples Synthesizing and Training for Robust Visual Question Answering
Long Chen, Yuhang Zheng, Yulei Niu, Hanwang Zhang, Jun Xiao
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 CSST-L contrastive training | UpDn · VQA-CP v2 test | accuracy (All): 39.68 -> 56.55 | Improved |
| add CSST-G contrastive training | LMH · VQA-CP v2 test | accuracy (All): 52.45 -> 61.66 | Improved |
| use CSST-L instead of CSST-G (ablation) | LMH · VQA-CP v2 test | accuracy (All): 61.66 -> 55.38 | Hurt |
| add CSST-G contrastive training | RUBi · VQA-CP v2 test | accuracy (All): 45.23 -> 47.06 | Improved |
| add CSST contrastive loss on top of CSS+ (ablation) | LMH · VQA-CP v2 test | accuracy (All): 59.54 -> 61.66 | Improved |
| add CSST contrastive loss on top of CSS+ (ablation) | LMH (LXMERT) · VQA-CP v2 test | accuracy (All): 63.63 -> 65.71 | Improved |
| add CSS+ synthesized samples (ablation) | LMH · VQA-CP v2 test | accuracy (All): 52.45 -> 59.54 | Improved |
| add improved CSS+ in place of original CSS (ablation) | LMH · VQA-CP v2 test | accuracy (All): 58.95 -> 59.54 | Improved |
| add SAR reranking | LMH-CSST · VQA-CP v1 test | accuracy (All): 63.16 -> 69.57 | Improved |
| add CSST-style CSS-negative selection instead of random-only negatives (ablation) | LMH CR · VQA-CP v2 test | accuracy (All): 60.74 -> 61.66 | Improved |
| add CSST selection in place of prior CL strategy (ablation) | LMH-CSS+ · VQA-CP v2 test | accuracy (All): 59.78 -> 61.66 | Improved |
| add CSS+ synthesized samples (ablation) | RUBi · GQA-OOD testdev | Delta (imbalance): 9.2 -> 14.5 | Mixed |
Closest papers by meaning
- Counterfactual Samples Synthesizing for Robust Visual Question Answering (CVPR 2020 · similarity 0.97)
- Contrast and Classify: Training Robust VQA Models (ICCV 2021 · similarity 0.91)
- Roses Are Red, Violets Are Blue... but Should VQA Expect Them To? (CVPR 2021 · similarity 0.91)
- Robust Visual Question Answering: Datasets, Methods, and Future Challenges (TPAMI 2024 · similarity 0.90)
- Counterfactual VQA: A Cause-Effect Look at Language Bias (CVPR 2021 · similarity 0.90)
- 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.
