PRISM-Net: Dynamic Routing Network with VLM-Guided Disentangled Attribute Distillation for Explainable Fine-Grained Product Retrieval
Authors: Chung-Ming Lo, Che-Wei Chang
Organizations: Institute of Artificial Intelligence Innovation, Industry Academia Innovation School, National Yang Ming Chiao Tung University, Hsinchu, Taiwan · Graduate Institute of Library, Information and Archival Studies, National Chengchi University, Taipei, Taiwan
Fine-grained visual product retrieval in large-scale e-commerce multimedia systems demands discriminative representations capable of simultaneously resolving macro-level category semantics and micro-level visual attributes such as structural contour (line) and chromatic appearance (color). However, conventional backbones including convolutional neural networks (CNNs) and vision Transformers (ViTs) inherently suffer from feature entanglement and shortcut learning. To learn the micro-level visual attributes without costly manual attribute annotations, we propose phase-routed interpretable semantic mixture-of-experts network (PRISM-Net), an explainable multi-expert architecture that disentangles holistic visual features into disentangled line and color expert pathways governed by a conditional meta-router. In phase 1, we introduce an external VLM-Guided sigmoid semantic distillation mechanism with gradient-detached routing gates, utilizing zero-shot text prompts from SigLIP to supervise specialized structural and chromatic experts. In phase 2, the disentangled expert embeddings are adaptively fused with macro base features via a learned meta-router gate to calibrate category boundaries. Extensive 5-fold cross-validation across 275,454 e-commerce images (18 categories) and full-scale retrieval benchmarks (55,091 queries against a 220,363-image gallery) across SwinV2, ConvNeXt-V2, and DINOv3 demonstrate substantial gains. On SwinV2, PRISM-Net boosts full-fold product mAP@10 from 80.98% to 83.90% (+2.92%), line mAP@10 by +5.74%, and color mAP@10 by +9.12%. Furthermore, unsupervised distribution analysis of the meta-router gate reveals that both ViT and CNN backbones emergently allocate up to 6.8x higher structural line weights to rigid appliances and hardware while dedicating >96% chromatic weight to clothing.
Existing approaches to visual attribute value extraction (AVE) primarily rely on static product images, failing to capture temporal cues, multi-angle views and fine-grained visual details. Directly applying video vision-language models (VLMs) to product AVE results in limited performance due to the lack of domain knowledge, and fine-tuning them requires extensive high-quality data and substantial computational resources. Thus, we propose visual search augmented chain-of-thought reasoning (ViS-CoT), a training-free, plug-and-play pipeline that can be easily applied to any open-source video VLM for video-to-text AVE in e-Commerce. Specifically, ViS-CoT employs visual clustering to identify representative frames, followed by visual search to retrieve semantically similar product knowledge that can enrich attribute cues. Next, an interleaved CoT reasoning module iteratively refines reasoning through visually-aligned auxiliary texts derived from captioning and automatic speech recognition. Finally, the integrated information guides the model toward accurate and fine-grained attribute predictions. Extensive experiments across 14 product categories on the VideoAVE dataset show that ViS-CoT consistently enhances multiple state-of-the-art video VLMs, achieving an average improvement of 17.91 percentage points in micro-F1.
Unifying the complementary strengths of diverse Vision Foundation Models (VFMs) into a single efficient model is highly desirable but challenged by the negative transfer inherent in monolithic distillation. To address these feature conflicts, we introduce PRISM, a novel dual-stream Mixture-of-Experts (MoE) framework that synergizes VFMs via modular specialization. We propose a two-stage paradigm: (1) expertise deconstruction, where a teacher-conditional router guides experts to specialize in distinct representational subspaces to mitigate interference, followed by (2) dynamic recomposition, where the router learns to assemble these experts into tailored computational pathways for downstream tasks. Experiments on PASCAL-Context and NYUD-v2 show that PRISM establishes a new state of the art, validating that sparse, emergent specialization is a scalable approach for integrating diverse visual knowledge.
Ying Tang, Dong Li, Youjia Zhang +3
School of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan, China.
Multimodal representation is crucial for E-commerce tasks such as identical product retrieval. Large representation models (e.g., VLM2Vec) demonstrate strong multimodal understanding capabilities, yet they struggle with fine-grained semantic comprehension, which is essential for distinguishing highly similar items. To address this, we propose Attribute-Enhanced Fine-Grained Multi-Modal Representation Learning (AFMRL), which defines product fine-grained understanding as an attribute generation task. It leverages the generative power of Multimodal Large Language Models (MLLMs) to extract key attributes from product images and text, and enhances representation learning through a two-stage training framework: 1) Attribute-Guided Contrastive Learning (AGCL), where the key attributes generated by the MLLM are used in the image-text contrastive learning training process to identify hard samples and filter out noisy false negatives. 2) Retrieval-aware Attribute Reinforcement (RAR), where the improved retrieval performance of the representation model post-attribute integration serves as a reward signal to enhance MLLM's attribute generation during multimodal fine-tuning. Extensive experiments on large-scale E-commerce datasets demonstrate that our method achieves state-of-the-art performance on multiple downstream retrieval tasks, validating the effectiveness of harnessing generative models to advance fine-grained representation learning.