This paper presents ECO, an efficient convolution operator for tracking, which addresses the issues of computational complexity and over-fitting in state-of-the-art Discriminative Correlation Filter (DCF) trackers. The key contributions of ECO include a factorized convolution operator that reduces the number of parameters in the model, a compact generative model of the training sample distribution that reduces memory and time complexity while enhancing sample diversity, and an efficient model update strategy that improves tracking speed and robustness. ECO achieves significant improvements in both performance and speed compared to the state-of-the-art DCF tracker C-COT. When using expensive deep features, ECO provides a 20-fold speedup and achieves a 13.0% relative gain in Expected Average Overlap (EAO) compared to C-COT on the VOT2016 benchmark. Additionally, a fast variant of ECO using hand-crafted features operates at 60 Hz on a single CPU and achieves a 65.0% AUC on OTB-2015. The ECO approach is evaluated on four benchmarks: VOT2016, UAV123, OTB-2015, and TempleColor. The results show that ECO significantly improves tracking performance and speed, achieving state-of-the-art results on all benchmarks. The factorized convolution operator reduces the number of parameters by 80%, training samples by 90%, and optimization iterations by 80% compared to the baseline. The compact generative model of the training sample distribution reduces memory and time complexity while maintaining sample diversity. The efficient model update strategy reduces over-fitting to recent samples and improves tracking robustness. The ECO approach is particularly effective in handling challenging tracking scenarios, such as scale variations, deformations, and out-of-plane rotations. The factorized convolution operator and compact generative model of the training sample distribution are key to achieving the improved performance and speed. The efficient model update strategy further enhances the robustness of the tracker by reducing over-fitting to recent samples. Overall, ECO provides a significant improvement in tracking performance and speed, making it a promising approach for real-time visual tracking applications.This paper presents ECO, an efficient convolution operator for tracking, which addresses the issues of computational complexity and over-fitting in state-of-the-art Discriminative Correlation Filter (DCF) trackers. The key contributions of ECO include a factorized convolution operator that reduces the number of parameters in the model, a compact generative model of the training sample distribution that reduces memory and time complexity while enhancing sample diversity, and an efficient model update strategy that improves tracking speed and robustness. ECO achieves significant improvements in both performance and speed compared to the state-of-the-art DCF tracker C-COT. When using expensive deep features, ECO provides a 20-fold speedup and achieves a 13.0% relative gain in Expected Average Overlap (EAO) compared to C-COT on the VOT2016 benchmark. Additionally, a fast variant of ECO using hand-crafted features operates at 60 Hz on a single CPU and achieves a 65.0% AUC on OTB-2015. The ECO approach is evaluated on four benchmarks: VOT2016, UAV123, OTB-2015, and TempleColor. The results show that ECO significantly improves tracking performance and speed, achieving state-of-the-art results on all benchmarks. The factorized convolution operator reduces the number of parameters by 80%, training samples by 90%, and optimization iterations by 80% compared to the baseline. The compact generative model of the training sample distribution reduces memory and time complexity while maintaining sample diversity. The efficient model update strategy reduces over-fitting to recent samples and improves tracking robustness. The ECO approach is particularly effective in handling challenging tracking scenarios, such as scale variations, deformations, and out-of-plane rotations. The factorized convolution operator and compact generative model of the training sample distribution are key to achieving the improved performance and speed. The efficient model update strategy further enhances the robustness of the tracker by reducing over-fitting to recent samples. Overall, ECO provides a significant improvement in tracking performance and speed, making it a promising approach for real-time visual tracking applications.