Research map / Vision-language and multimodal / Human and embodied modeling
Learning Better Visual Dialog Agents With Pretrained Visual-Linguistic Representation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use pretrained VilBERT with two-way background/target fusion | Oracle · GuessWhat?! test | independent accuracy: 78.5% -> 85.0% | Helped |
| add Faster-RCNN visual features to baseline Oracle | Oracle · GuessWhat?! test | independent accuracy: 78.5% -> 81.7% | Helped |
| use pretrained VilBERT and turn-level state tracking | Guesser · GuessWhat?! test | independent accuracy: 69.5% -> 76.5% | Helped |
| replace baseline Questioner with VilBERT-Questioner | Baseline Oracle + Baseline Guesser self-play · GuessWhat?! test | success rate: 45.9% over VDST; 52.5% | Helped |
| use all three proposed VilBERT agents together | GuessWhat?! self-play · GuessWhat?! test | success rate: 44.6% -> 62.8% | Helped |
| use a pretrained VilBERT-Guesser state-estimator | VilBERT-Questioner · GuessWhat?! test | games with repeated questions: 40.05 -> 32.56 | Helped |
| inject answer information into post-layers rather than concatenate question-answer input (ablation) | VilBERT-Guesser · GuessWhat?! test | independent accuracy: 74.3% -> 76.5% | Helped |
| feed turn-level questions rather than concatenate the entire dialog history (ablation) | VilBERT-Guesser · GuessWhat?! test | independent accuracy: 65.7% -> 76.5% | Helped |
| freeze rather than fine-tune the pretrained state-estimator (ablation) | VilBERT-Questioner · GuessWhat?! test | success rate: 57.0% -> 62.8% | Helped |
| replace VilBERT-Oracle with Baseline Oracle (ablation) | VilBERT-Guesser + VilBERT-Questioner self-play · GuessWhat?! test | success rate: 62.8% -> 52.7% | Hurt |
| use VilBERT-Oracle and VilBERT-Guesser together instead of only one proposed agent (ablation) | self-play with VDST · GuessWhat?! test | success rate: 50.6% -> 55.7% | Helped |
| replace Baseline Oracle with a BERT-language-encoder Baseline Oracle (ablation) | Baseline Oracle · GuessWhat?! test | accuracy: 78.67% versus 85.0% | No clear effect |
Closest papers by meaning
- Visual Dialog (CVPR 2017 · similarity 0.91)
- Divide-and-Conquer: Tree-structured Strategy with Answer Distribution Estimator for Goal-Oriented Visual Dialogue (AAAI 2025 · similarity 0.89)
- Automated Generation of Challenging Multiple-Choice Questions for Vision Language Model Evaluation (CVPR 2025 · similarity 0.87)
- Tackling Vision Language Tasks through Learning Inner Monologues (AAAI 2024 · similarity 0.87)
- Structure-Aware Multimodal Sequential Learning for Visual Dialog (AAAI 2024 · similarity 0.86)
- Cross-modal Information Flow in Multimodal Large Language Models (CVPR 2025 · similarity 0.86)
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.