[Submitted on 14 Apr 2026 (v1), last revised 15 Oct 2026 (this version, v2)]
Abstract:Multimodal large language models (MLLMs) perform well on many vision-language tasks but often struggle with vision-centric problems that require fine-grained visual perception. Recent evidence suggests that this limitation arises not from weak visual representations, but from under-utilization of visual information during instruction tuning, where many tasks can be partially solved using language priors alone. We propose a simple and lightweight approach that augments visual instruction tuning with a small number of visually grounded self-supervised tasks expressed as natural language instructions. By reformulating classical self-supervised pretext tasks, such as rotation prediction, color matching, and cross-view correspondence, as image-instruction-response triplets, we introduce supervision that cannot be solved without relying on visual evidence. Our approach requires no human annotations, no architectural modifications, and no additional training stages. Across multiple models, training regimes, and benchmarks, injecting only a small fraction (3-10%) of such visually grounded instructions consistently improves performance on vision-centric evaluations. Our findings highlight instruction tuning with visually grounded SSL tasks as a powerful lever for improving visual perception in MLLMs through simple adjustments to the training data distribution. Code available at: this https URL
Submission history
From: Sophia Sirko-Galouchenko [view email]
[v1] Tue, 14 Apr 2026 16:59:53 UTC (7,679 KB)
[v2] Thu, 15 Oct 2026 11:20:06 UTC (8,124 KB)
