Multi-Scale Factor Super-Resolution for Light Field Images

Light field image super-resolution (LFSR) has achieved significant progress with the rapid development of deep learning techniques. However, most existing methods are limited to a single scale factor, requiring the training and storage of separate models for each upsampling task (e.g., 2×/4× spatial SR and 7×7/8×8 angular SR), which is inefficient for deployment. In this paper, we propose a unified end-to-end framework, termed as MLFSR, which can handle multi-scale factor LFSR tasks within a single model. The MLFR consists of three modules: the light field feature disentangling module (LFDM), which effectively separates and fuses spatial and angular features; the scale-aware feature adaptation module (SFAM), which enables the network to adaptively process feature representations for different scale factors; and the multi-scale factor upsampling module (MFUM), which utilizes task-specific sub-modules for efficient high-resolution reconstruction. Extensive experiments on both synthetic and real-world datasets demonstrate that our method achieves competitive performance compared to state-of-the-art methods while maintaining a single model for all tested scale factors.