Research map / Vision-language and multimodal / Multimodal LLMs
"X-InstructBLIP: A Framework for Aligning Image, 3D, Audio, Video to LLMs and its Emergent Cross-modal Reasoning"
Artemis Panagopoulou, Le Xue, Ning Yu, LI JUNNAN, DONGXU LI, Shafiq Joty, Ran Xu, Silvio Savarese, Caiming Xiong, Juan Carlos Niebles
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 modality-prefix tokens to LLM input (ablation) | X-Instruct Proj. (7b) · ESC50 closed | accuracy: 67.5 -> 75.9 | Improved |
| add modality-prefix tokens (ablation) | X-Instruct Proj. (7b) · MusicAVQA audio+video delta | joint-minus-best-single: -5.0 -> 1.3 | Improved |
| add modality-prefix tokens (ablation) | X-Instruct Proj. (7b) · DisCRn Image-3D | accuracy: 57.7 -> 48.1 | Hurt |
| initialize audio Q-Former from BLIP-2 instead of random (ablation) | X-Instruct Proj. (7b) · ESC50 closed | accuracy: 70.0 -> 75.9 | Improved |
| BLIP-2 initialization on ClothoAQA (ablation) | X-Instruct Proj. (7b) · ClothoAQA | accuracy: 11.9 -> 15.4 | Improved |
| use instruction-aware (X-Instruct) projection instead of LLaVA-style projection | X-InstructBLIP · ModelNet40 closed | accuracy: 46.4 -> 62.8 | Improved |
| use instruction-aware projection instead of LLaVA-style on audio-video joint reasoning | X-InstructBLIP · VATEX joint audio+video | joint-minus-best-single: -10.7 -> 1.7 | Improved |
| X-InstructBLIP versus InstructBLIP on general image tasks | X-InstructBLIP · Flickr30k, NoCaps, MME | CIDEr / MME: 84.5 -> 82.1 Flickr30k | Mixed |
| LLaVA-style versus X-Instruct projection on DisCRn audio-video | X-InstructBLIP · DisCRn Audio-Video | accuracy: 47.1 vs 34.0 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Pengi — Pengi: An Audio Language Model for Audio Tasks (NeurIPS 2023) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- XModBench: Benchmarking Cross-Modal Capabilities and Consistency in Omni-Language Models (ICLR 2026 · similarity 0.91)
- Evaluating Cross-Modal Reasoning Ability and Problem Charactaristics with Multimodal Item Response Theory (ICLR 2026 · similarity 0.91)
- ROVER: Benchmarking Reciprocal Cross-Modal Reasoning for Omnimodal Generation (ICLR 2026 · similarity 0.91)
- CREMA: Generalizable and Efficient Video-Language Reasoning via Multimodal Modular Fusion (ICLR 2025 · similarity 0.90)
- Measuring Cross-Modal Synergy: A Benchmark for VLM Explainability (NeurIPS 2026 · similarity 0.90)
- CRIT: Graph-Based Automatic Data Synthesis to Enhance Cross-Modal Multi-Hop Reasoning (CVPR 2026 · 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.
