PFGNet：一种用于高效时空预测学习的全卷积频率引导外周门控网络
PFGNet: A Fully Convolutional Frequency-Guided Peripheral Gating Network for Efficient Spatiotemporal Predictive Learning
用像素级频率引导门控动态调制感受野、以可分离一维大核卷积替代循环/注意力做时空建模，为视频预测与时空表征提供了一种全并行、低 FLOPs 的卷积骨干设计思路。
原文依据 dynamically modulates receptive fields through pixel-wise frequency-guided gating
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论时空预测学习（STPL）旨在根据过去的观测预测未来帧，在广泛的应用中至关重要。与循环或混合架构相比，纯卷积模型具有更优的效率和完全并行性，但其固定的感受野限制了其自适应捕捉空间变化运动模式的能力。受生物中心-外周组织与频率选择性信号处理的启发，我们提出 PFGNet，一个通过像素级频率引导门控动态调制感受野的全卷积框架。核心的外周频率门控（PFG）模块提取局部频谱线索，并自适应地将多尺度大核外周响应与可学习的中心抑制相融合，从而有效形成空间自适应的带通滤波器。为保持效率，所有大核均分解为可分离的一维卷积（先 1xk 再 kx1），将每通道计算开销从 O(k2) 降至 O(2k)。PFGNet 无需循环或注意力即可实现结构感知的时空建模。在 Moving MNIST、TaxiBJ、Human3.6M 和 KTH 上的实验表明，PFGNet 以显著更少的参数和 FLOPs 取得了 SOTA 或接近 SOTA 的预测性能。代码已开源：https://github.com/fhjdqaq/PFGNet。
英文原文摘要 1349 字符
Spatiotemporal predictive learning (STPL) aims to forecast future frames from past observations and is essential across a wide range of applications. Compared with recurrent or hybrid architectures, pure convolutional models offer superior efficiency and full parallelism, yet their fixed receptive fields limit their ability to adaptively capture spatially varying motion patterns. Inspired by biological center-surround organization and frequency-selective signal processing, we propose PFGNet, a fully convolutional framework that dynamically modulates receptive fields through pixel-wise frequency-guided gating. The core Peripheral Frequency Gating (PFG) block extracts localized spectral cues and adaptively fuses multi-scale large-kernel peripheral responses with learnable center suppression, effectively forming spatially adaptive band-pass filters. To maintain efficiency, all large kernels are decomposed into separable 1D convolutions (1xk followed by kx1), reducing per-channel computational cost from O(k2) to O(2k). PFGNet enables structure-aware spatiotemporal modeling without recurrence or attention. Experiments on Moving MNIST, TaxiBJ, Human3.6M, and KTH show that PFGNet delivers SOTA or near-SOTA forecasting performance with substantially fewer parameters and FLOPs. Our code is available at https://github.com/fhjdqaq/PFGNet.