Search papers, labs, and topics across Lattice.
The paper introduces SWIM, a training strategy to align vision and language representations in multimodal LLMs for fine-grained object understanding from text prompts alone. It addresses a discovered misalignment where object nouns yield diffuse visual activations compared to attribute words. SWIM uses a new dataset, NL-Refer, to enforce spatial consistency between object noun cross-attention maps and ground-truth masks during training, improving text-visual alignment and performance.
Multimodal LLMs struggle to pinpoint objects from nouns alone, but SWIM training realigns vision and language to outperform visual-prompt methods.
We present SWIM (See What I Mean), a novel training strategy that aligns vision and language representations to enable fine-grained object understanding solely from textual prompts. Unlike existing approaches that require explicit visual prompts, such as masks or points, SWIM leverages mask supervision only during training to guide cross-modal attention, allowing the model to automatically attend to the user-specified object at inference. Our cross-attention analysis of pretrained multimodal large languagemodels (MLLMs) reveals a systematic discrepancy: Attribute words produce sharp, localized activations in the visual modality, whereas object nouns yield diffuse and scattered patterns due to semantic reference bias and distributed high-level representations. To address this misalignment, we construct NL-Refer, an enriched dataset, in which each object mask is paired with a precise natural language referring expression. SWIM extracts multi-layer cross-attention maps from object nouns and enforces spatial consistency with ground-truth masks. Experimental results demonstrate that SWIM substantially improves text-visual alignment and achieves superior performance over visual-prompt-based methods on fine-grained object understanding benchmarks. The code and data are available at https://github.com/HumanMLLM/SWIM{https://github.com/HumanMLLM/SWIM}.
DAMO