# Crab+ Unifies AV-LLMs, Reverses Negative Transfer

_Crab+ introduces a novel approach to Audio-Visual Large Language Models, overcoming negative transfer via explicit cooperation in data and model design._

**Published:** 2026-03-05
**Source:** https://www.startuphub.ai/ai-news/ai-research/2026/crab-unifies-av-llms-reverses-negative-transfer

---

The pursuit of unified scene understanding in [multimodal intelligence](/ai-news/ai-research/2026/multimodal-llms-what-s-lost-in-translation) hinges on effective Audio-Visual Large Language Models (AV-LLMs). However, conventional multi-task learning approaches for these models are plagued by substantial negative transfer, degrading performance on nearly 55% of tasks compared to single-task training. This stems from the inherent heterogeneity of audio-visual tasks, marked by varying granularity and distinct capability demands, leading to interference during joint training.

## Tackling Task Heterogeneity with Explicit Cooperation

To overcome this critical bottleneck, the researchers introduce Crab$^{+}$, a scalable model designed for unified audio-visual scene understanding. Crab$^{+}$ tackles task heterogeneity through explicit cooperation at both the data and model levels. On the data front, they present AV-UIE v2, a comprehensive dataset featuring approximately 222K samples across 17 datasets and 7 tasks, augmented with explicit reasoning processes. This dataset enables the model to better capture cross-task relationships at diverse granularities. Complementing this, the model architecture features a unified interface to standardize task formulations and introduces Interaction-aware LoRA (I-LoRA). I-LoRA employs dynamic routing to explicitly model inter-task relationships, thereby coordinating distinct audio-visual interaction patterns and mitigating parameter interference.

## Demonstrating Positive Transfer and Broad Applicability

The experimental results from [Crab$^{+}$](https://arxiv.org/abs/2603.04128v1) are compelling. The model not only covers a broader range of tasks than existing unified AV-LLMs but also surpasses specialized models on various benchmarks. Crucially, Crab$^{+}$ reverses the negative transfer trend, achieving positive transfer where multi-task learning now outperforms single-task baselines in nearly 88% of evaluated tasks. These gains are consistent across diverse AV-LLM paradigms and are validated by in-depth visualizations, positioning Crab$^{+}$ as a robust advancement in holistic audio-visual scene understanding.

---

Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory.