GRNN:Recurrent Neural Network based on Ghost Features for Video Super-Resolution

TL;DR

This work addresses the computational burden and feature redundancy in video super-resolution by introducing GRNN, a recurrent framework that incorporates Ghost features to reduce redundant feature maps. By coupling Ghost blocks with a residual RNN structure, the approach mitigates gradient disappearance and preserves texture over longer sequences. Across Vimeo-90k training and standard VSR benchmarks (Vid4, SPMCS, UDM10), GRNN demonstrates improved PSNR and SSIM and delivers visually sharper textures, highlighting the practical impact of efficient temporal modeling in VSR.

Abstract

Modern video super-resolution (VSR) systems based on convolutional neural networks (CNNs) require huge computational costs. The problem of feature redundancy is present in most models in many domains, but is rarely discussed in VSR. We experimentally observe that many features in VSR models are also similar to each other, so we propose to use "Ghost features" to reduce this redundancy. We also analyze the so-called "gradient disappearance" phenomenon generated by the conventional recurrent convolutional network (RNN) model, and combine the Ghost module with RNN to complete the modeling on time series. The current frame is used as input to the model together with the next frame, the output of the previous frame and the hidden state. Extensive experiments on several benchmark models and datasets show that the PSNR and SSIM of our proposed modality are improved to some extent. Some texture details in the video are also better preserved.

Figures (7)

• Figure 1: Similar feature maps where the rectangular boxes of the same color are similar to each other
• Figure 2: Flowchart of generating Ghost Block
• Figure 3: Flowchart of RNN
• Figure 4: Relevant code module of RNN
• Figure 5: GRNN Model Structure. The box (a) shows the overall design idea of the model, and the internal structure of Ghost Block is the blue box (b).