GhostSR: Learning Ghost Features for Efficient Image Super-Resolution
This work addresses efficiency issues in SISR for applications requiring real-time or resource-constrained image enhancement, though it is incremental as it builds on existing CNN-based SISR methods.
The paper tackles the problem of feature redundancy in single image super-resolution (SISR) models, which leads to high computational costs, by proposing GhostSR, a method that uses learnable shift operations to generate redundant features, achieving comparable performance to baselines with significant reductions in parameters, FLOPs, and GPU inference latency, such as a 46% reduction in parameters and FLOPs and a 42% reduction in latency for a ×2 EDSR network.
Modern single image super-resolution (SISR) system based on convolutional neural networks (CNNs) achieves fancy performance while requires huge computational costs. The problem on feature redundancy is well studied in visual recognition task, but rarely discussed in SISR. Based on the observation that many features in SISR models are also similar to each other, we propose to use shift operation to generate the redundant features (i.e., ghost features). Compared with depth-wise convolution which is time-consuming on GPU-like devices, shift operation can bring a practical inference acceleration for CNNs on common hardwares. We analyze the benefits of shift operation on SISR task and make the shift orientation learnable based on Gumbel-Softmax trick. Besides, a clustering procedure is explored based on pre-trained models to identify the intrinsic filters for generating intrinsic features. The ghost features will be derived by moving these intrinsic features along a specific orientation. Finally, the complete output features are constructed by concatenating the intrinsic and ghost features together. Extensive experiments on several benchmark models and datasets demonstrate that both the non-compact and lightweight SISR models embedded with the proposed method can achieve a comparable performance to that of their baselines with a large reduction of parameters, FLOPs and GPU inference latency. For instance, we reduce the parameters by 46%, FLOPs by 46% and GPU inference latency by 42% of $\times2$ EDSR network with basically lossless performance.