Research map / Vision-language and multimodal / Multimodal LLMs
BASIC: Boosting Visual Alignment with Intrinsic Refined Embeddings in Multimodal Large Language Models
Jianting Tang, Yubo Wang, Haoyu Cao, Linli Xu
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 directional alignment loss L_das (ablation) | BASIC-7B (Vicuna-7B) · MMBench-EN | score: 64.3 -> 68.6 | Improved |
| add semantic distribution loss L_sds (ablation) | BASIC-7B (Vicuna-7B) · MMBench-EN | score: 64.3 -> 68.0 | Improved |
| add both L_das and L_sds (ablation) | BASIC-7B (Vicuna-7B) · ScienceQA-IMG | score: 66.8 -> 70.6 | Improved |
| add both L_das and L_sds over either alone (ablation) | BASIC-7B (Vicuna-7B) · MM-Vet | score: 33.1 -> 33.8 | Improved |
| add both losses | BASIC-7B (Vicuna-7B) · TextVQA | score: 58.2 -> 58.0 | Hurt |
| add both losses (ablation) | BASIC-7B (Vicuna-7B) · TextVQA | score: 57.7 -> 58.0 | Hurt |
| increasing layer weights instead of decreasing (layers 1-16) (ablation) | BASIC-7B (Vicuna-7B) · MM-Vet | score: 27.8 -> 28.7 | Improved |
| increasing layer weights instead of constant (ablation) | BASIC-7B (Vicuna-7B) · GQA | score: 55.6 -> 55.9 | Improved |
| importance-based supervision instead of equal supervision (ablation) | BASIC-7B (Vicuna-7B) · MM-Vet | score: 27.6 -> 28.7 | Improved |
| importance-based supervision instead of equal supervision (ablation) | BASIC-7B (Vicuna-7B) · VQAv2 | score: 73.0 -> 73.4 | Improved |
| use refined embeddings from deep layers (16-32) in supervision (ablation) | BASIC-7B (Vicuna-7B) · benchmark average | score | Hurt |
| apply BASIC to Gemma-2B | Gemma-2B · MMBench-CN | score: 49.5 -> 51.2 | Improved |
Later papers that compared with this method
1 later accepted paper reports BASIC as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Learning Sample-wise Rank-Aware Interpolation Weights for Composed Visual Data Retrieval (ECCV 2026 · better on 1 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
- Analyzing Fine-Grained Alignment and Enhancing Vision Understanding in Multimodal Language Models (NeurIPS 2025 · similarity 0.94)
- UVU: Improving Multimodal Understanding via Vision-Language Unified Autoregressive Paradigm (CVPR 2026 · similarity 0.92)
- VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks (ICLR 2025 · similarity 0.92)
- AlignVLM: Bridging Vision and Language Latent Spaces for Multimodal Document Understanding (NeurIPS 2025 · similarity 0.91)
- Unleashing the Intrinsic Visual Representation Capability of Multimodal Large Language Models (CVPR 2026 · similarity 0.91)
- Bridging Vision and Language Spaces with Assignment Prediction (ICLR 2024 · similarity 0.91)
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.
