ResFlow: Fine-Tuning Residual Optical Flow for Event-Based High Temporal Resolution Motion Estimation

Event cameras offer great potential for high-temporal-resolution (HTR) motion estimation in dynamic real-world scenarios. However, the lack of dense HTR ground truth in real-world datasets prevents fully leveraging the high temporal resolution potential of event cameras. Furthermore, the intrinsic sparsity of event data introduces additional challenges for optimization and supervision. To address these issues, we propose a residual-based paradigm that decomposes HTR optical flow into a global linear component and high-frequency residuals. The residual paradigm effectively mitigates the impacts of event sparsity on optimization and is compatible with any LTR algorithm. In addition, to bridge the supervision gap caused by the lack of HTR ground truth, we incorporate novel learning strategies. Specifically, we initially employ a shared refiner to estimate the residual flows, enabling both LTR supervision and HTR inference. Subsequently, we introduce regional noise to simulate the residual patterns of intermediate flows, facilitating the adaptation from LTR supervision to HTR inference. Additionally, we show that the noise-based strategy supports in-domain self-supervised training. Comprehensive experimental results demonstrate that our approach achieves state-of-the-art accuracy among existing HTR methods, highlighting its effectiveness and superiority. The source code will be publicly available at https://github.com/ZhouQianang/ResFlow.