Search papers, labs, and topics across Lattice.
This paper introduces SE-MoLoRA, a parameter-efficient framework that enhances domain-specific photographic assessment by disentangling general photographic knowledge from specialized critiques. By utilizing a shared LoRA expert alongside routed adapters for composition, lighting, and technical quality, the method enables targeted critique without the need for training separate full models. The results show a significant improvement in critique generation performance, with BERTScore-F1 increasing from 0.2317 to 0.4215 and a preference rate of 84.6% in pairwise comparisons against monolithic LoRA models.
Modular adaptation in SE-MoLoRA not only boosts critique performance but also streamlines the process, reducing the need for extensive model training.
Vision-language models can describe images fluently, but they often fail to provide actionable photographic critique because semantic content and aesthetic judgment remain entangled. We propose SE-MoLoRA, a modular parameter-efficient adaptation framework for domain-specific photographic assessment. The method separates general photographic knowledge from specialist residual judgments using an always-active shared LoRA expert and routed adapters for composition, lighting, and technical quality. A lightweight query router selects the relevant specialist, enabling targeted critique without training separate full models. A rank-64 shared adapter captures broad photographic vocabulary, while rank-32 specialists learn domain-specific residuals with an orthogonal regularization penalty that encourages disentangled representations. Training data is obtained by distilling the Reddit Photo Critique Dataset into domain-labeled critique samples. On held-out critique generation, SE-MoLoRA improves BERTScore-F1 from 0.2317 to 0.4215 over monolithic LoRA and is preferred in 84.6\% of pairwise comparisons, while using fewer active parameters than separate specialist models. SVD-based ablation study shows that shared-specialist decomposition and orthogonal regularization reduce expert overlap. These results demonstrate that modular adaptation improves controllability and specificity in multimodal photographic critique.
