Research map / Vision-language and multimodal / Visual question answering
Perceptual Score: What Data Modalities Does Your Model Perceive?
Itai Gat, Idan Schwartz, Alex Schwing
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 |
|---|---|---|---|
| train LMH on CSS-generated counterfactual samples with unmodified inputs (ablation) | LMH · VQA-CP | accuracy: 57.54% (vs CSS 57.89, LMH 52.23) | Improved |
| CSS counterfactual data shifts yes/no prior toward test distribution | CSS · VQA-CP Yes/No | image perceptual score P_V/Z_D: 0.71 vs LMH 20.58 | Hurt |
| use answer-only features instead of all SocialIQ modalities (ablation) | SocialIQ baseline · SocialIQ val | accuracy: 64.84 -> 68.65 | Improved |
| change baseline to FGA with GloVe features | SocialIQ baseline · SocialIQ val | accuracy: 64.84 -> 67.38 | Improved |
| untrained NLTK sentiment classifier on answer text | NLTK-Sentiment · SocialIQ val | accuracy: 66.7% vs baseline 64.84 | Improved |
| lower image quality in audio-visual crowd counting | audio-visual model · DISCO | audio perceptual score P_A/Z_D: 9.22 -> 23.91 | Improved |
| compare ranking metric NDCG vs MRR for question perception | LS vs FGA · VisDial v1.0 | question perceptual score P_Q/Z_{D,f}: MRR: FGA 51.26 vs LS 30.49; NDCG: LS 63.14 vs FGA 45.11 | Mixed |
| fine-tune LS on dense annotations (LS (CE)) vs LS | LS · VisDial v1.0 MRR | MRR: 69.00 -> 52.21 | Hurt |
| use caption modality in dialog | LS, FGA, LS (CE) · VisDial v1.0 | caption perceptual score: 0.00 to 1.83 | No clear effect |
| model-normalize rather than task-normalize perceptual score for LMH (lowest accuracy) | LMH · VQAv2 | image perceptual score: 48.64 vs LXMERT 47.16 | No clear effect |
| use nonlinear network instead of logistic regression on synthetic data | perceptual score · Synthetic | P_a, P_b (avg): 0.39 -> 35.72 (P_a), 0.56 -> 36.8 (P_b) | Improved |
| increase variance of modality c from 0 to 1 | perceptual score · Synthetic | P_c: 0 -> 32.25 | Improved |
Closest papers by meaning
- Assessing Modality Bias in Video Question Answering Benchmarks with Multimodal Large Language Models (AAAI 2025 · similarity 0.84)
- Words or Vision: Do Vision-Language Models Have Blind Faith in Text? (CVPR 2025 · similarity 0.84)
- Are We on the Right Way for Evaluating Large Vision-Language Models? (NeurIPS 2024 · similarity 0.84)
- Q-Bench: A Benchmark for General-Purpose Foundation Models on Low-level Vision (ICLR 2024 · similarity 0.84)
- Q-Instruct: Improving Low-level Visual Abilities for Multi-modality Foundation Models (CVPR 2024 · similarity 0.84)
- Same Task, Different Circuits: Disentangling Modality-Specific Mechanisms in VLMs (NeurIPS 2025 · similarity 0.84)
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.
