| Title |
Lightweight Image Super-resolution with Progressive Filter Pruning and Adaptive Response Distillation |
| Authors |
박정혁(Jeong Hyeok Park) ; 송병철(Byung Cheol Song) |
| DOI |
https://doi.org/10.5573/ieie.2026.63.1.35 |
| Keywords |
Deep learning; Image super-resolution; Knowledge distillation; Network filter pruning |
| Abstract |
Image super resolution (SR), which reconstructs high resolution images from low resolution inputs, has achieved remarkable performance improvements with the advancement of deep learning. However, recent SR models based on transformers and diffusion models require substantial computational resources, increasing the need for model compression. To address this issue, lightweight SR methods using model pruning and knowledge distillation (KD) have been actively studied, and recent approaches combine these techniques to improve both efficiency and performance. In this paper, we propose a two stage training framework that sequentially applies model pruning and knowledge distillation for efficient SR model compression. First, we perform effective model pruning through progressive local filter pruning, which balances filter removal and structural preservation. Subsequently, we apply mask based adaptive response distillation, which improves upon conventional L1 response distillation by focusing on transferring critical response information. Experimental results on multiple benchmark SR datasets demonstrate that our framework achieves both higher performance and efficiency compared with existing lightweight SR models. |