LRConv-NeRV: Low-Rank Convolution for Efficient Neural Video Compression

Abstract

Neural Representations for Videos (NeRV) encode entire video sequences within neural network parameters, offering an alternative paradigm to conventional video codecs. However, the convolutional decoder of NeRV remains computationally expensive and memory intensive, limiting its deployment in resource-constrained environments. This paper proposes LRConv–NeRV, an efficient NeRV variant that replaces selected dense $3\times 3$ convolutional layers with structured low-rank separable convolutions, trained end-to-end within the decoder architecture. By progressively applying low-rank factorization from the largest to earlier decoder stages, LRConv-NeRV enables controllable trade-offs between reconstruction quality and efficiency. Unlike prior approaches that rely on numerical compression (e.g., binarization) or feature redundancy (e.g., Ghost convolution), the proposed LRConv–NeRV enforces a low-rank constraint directly within the convolutional operator during end-to-end training, enabling structured reduction of computational complexity while preserving representational capacity. Extensive experiments conducted on the Bunny sequence and additional sequences from the UVG dataset demonstrate that applying LRConv to the final decoder stage only reduces decoder complexity by 68% (from 201.9 to 64.9 GFLOPs) and decreases model size by 9.3%, while maintaining reconstruction quality with less than 0.5 dB PSNR degradation in the best-performing configurations. Furthermore, the proposed method achieves up to 9.2% bitrate reduction relative to the NeRV baseline under identical training, pruning, and quantization settings. Under INT8 post-training quantization, LRConv–NeRV preserves reconstruction quality close to the dense NeRV baseline, whereas more aggressive factorization of early decoder stages leads to disproportionate quality degradation. Compared to existing work under layer-aligned settings, LRConv–NeRV achieves a more favorable efficiency versus quality trade-off, offering substantial GFLOPs and parameter reductions while maintaining higher PSNR/MS-SSIM and improved temporal stability. Temporal flicker analysis using LPIPS further shows that the proposed solution preserves temporal coherence close to the NeRV baseline, results establish LRConv–NeRV as a potential architectural alternative for efficient neural video decoding under low-precision and resource-constrained settings.