Research map / Vision-language and multimodal / Human and embodied modeling
NLX-GPT: A Model for Natural Language Explanations in Vision and Vision-Language Tasks
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 |
|---|---|---|---|
| Add image-caption pretraining (ablation) | NLX-GPT · VQA-X unfiltered | BLEU-4: 23.8 -> 25.6 | Helped |
| Add image-caption pretraining (ablation) | NLX-GPT · ACT-X unfiltered | CIDER: 63.5 -> 74.9 | Helped |
| Add Visual Genome concept inputs (ablation) | NLX-GPT · e-SNLI-VE filtered | BLEU-4: 11.9 -> 12.9 | Helped |
| Add Visual Genome concept inputs (ablation) | NLX-GPT · e-SNLI-VE filtered | METEOR: 18.1 -> 18.8 | Helped |
| Add Visual Genome concept inputs (ablation) | NLX-GPT · e-SNLI-VE filtered | CIDER: 114.7 -> 117.4† | Mixed |
| Replace CLIP ViT with CLIP ResNet-101 (ablation) | NLX-GPT · VQA-X filtered, no image-caption pretraining | BLEU-4: 28.1 -> 26.7 | Hurt |
| Replace CLIP ViT with ImageNet-pretrained ResNet-101 (ablation) | NLX-GPT · VQA-X filtered, no image-caption pretraining | BLEU-4: 28.1 -> 22.4 | Hurt |
| Replace CLIP ViT with bottom-up features (ablation) | NLX-GPT · VQA-X filtered, no image-caption pretraining | BLEU-4: 28.1 -> 26.5 | Hurt |
| Model answer generation jointly with its explanation instead of using a separate task model | NLX-GPT · VQA-X | task accuracy: 3% | Helped |
| Use NLX-GPT instead of the strongest listed baseline | NLX-GPT · e-SNLI-VE filtered | METEOR: 19.6 -> 18.8 | Hurt |
| Use NLX-GPT instead of the strongest listed baseline | NLX-GPT · e-SNLI-VE filtered | BLEU-4: 9.6 -> 12.9 | Helped |
| Increase retrieval set size from K@5 to K@10 | NLX-GPT · VQA-X image-bias retrieval attack | intra-distance: 44.56 -> 44.99 | Helped |
Closest papers by meaning
- Natural Language Inference Improves Compositionality in Vision-Language Models (ICLR 2025 · similarity 0.89)
- Language Model as Visual Explainer (NeurIPS 2024 · similarity 0.89)
- Vision-Language Models Do Not Understand Negation (CVPR 2025 · similarity 0.89)
- Synthesize Diagnose and Optimize: Towards Fine-Grained Vision-Language Understanding (CVPR 2024 · similarity 0.88)
- NL-Eye: Abductive NLI For Images (ICLR 2025 · similarity 0.88)
- Revisiting the Role of Language Priors in Vision-Language Models (ICML 2024 · similarity 0.88)
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.