Research questionHow can neural video representation decoders reduce compute and memory use on resource-constrained devices without hurting reconstruction quality?Encoding an entire video in neural-network parameters makes decoding depend on a convolutional network whose compute and memory costs can be prohibitive on constrained hardware. Reducing those costs can affect reconstruction quality, bitrate, and temporal stability. Latest papersRecent research connected to this question, newest first.LRConv-NeRV: Low Rank Convolution for Efficient Neural Video CompressionThe source studies NeRV-style neural video representations with convolutional decoders, including INT8 post-training quantization. Reported evidence covers GFLOPs, model size, bitrate, PSNR, MS-SSIM, temporal stability, and LPIPS-based flicker, with greater quality degradation observed when compression is applied to earlier decoder stages.research paper · Sep 2, 2026