InternVQA: Advancing Compressed Video Quality Assessment with Distilling Large Foundation Model

2025-02-26

AI Summary

This work addresses the limited representational capacity of lightweight models in compressed video quality assessment. We propose a novel knowledge distillation paradigm leveraging the InternVideo2 large vision-language model, marking the first application of video foundation models to compression distortion modeling. Our approach introduces a compression-quality-prior-guided distillation strategy, incorporating multi-backbone architecture comparison, compression-domain feature alignment loss, and a lightweight quality regression head. Evaluated on mainstream compressed video quality datasets, the distilled lightweight model substantially outperforms conventional methods and existing lightweight competitors—reducing parameter count by 98.7% and accelerating inference by 3.2×. Key contributions are: (1) pioneering the adaptation of video foundation models to compression distortion perception; (2) designing a distillation framework explicitly aligned with compression-domain priors; and (3) systematically validating the efficacy of multi-backbone architectures in quality-aware distillation.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Learning on the Edge & Model Compression
• Data Mining & Knowledge Management: Data Compression

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Video quality assessment tasks rely heavily on the rich features required for video understanding, such as semantic information, texture, and temporal motion. The existing video foundational model, InternVideo2, has demonstrated strong potential in video understanding tasks due to its large parameter size and large-scale multimodal data pertaining. Building on this, we explored the transferability of InternVideo2 to video quality assessment under compression scenarios. To design a lightweight model suitable for this task, we proposed a distillation method to equip the smaller model with rich compression quality priors. Additionally, we examined the performance of different backbones during the distillation process. The results showed that, compared to other methods, our lightweight model distilled from InternVideo2 achieved excellent performance in compression video quality assessment.

Problem

• Enhance compressed video quality assessment
• Transfer large foundation model capabilities
• Develop lightweight model via distillation

Innovation

• Distillation of large model
• Lightweight compression quality assessment
• Transferability of InternVideo2