arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2207.01271cs.CV
收藏

FlowNAS:用于光流估计的神经结构搜索

FlowNAS: Neural Architecture Search for Optical Flow Estimation

Wangxuan Institute of Computer Technology, Peking University(北京大学王选计算机研究所);University of California, Merced(加州大学默塞德分校)

Zhiwei Lin, Tingting Liang, Taihong Xiao, Yongtao Wang, Zhi Tang, Ming-Hsuan Yang

摘要:

现有的光流估计器通常采用通常为图像分类设计的网络架构作为编码器来提取每像素特征。然而,由于任务之间的自然差异,为图像分类设计的架构对于流量估计可能是次优的。为了解决这个问题,我们提出了一种名为FlowNAS的神经架构搜索方法,以自动找到更好的编码器架构进行流估计任务。我们首先设计一个合适的搜索空间,包括各种卷积运算符,并构建一个权重共享的超级网络,有效地评估候选架构。然后,为了更好地训练超级网络,我们提出了特征对齐蒸馏,它利用经过良好训练的流量估计器来指导超级网络的训练。最后,利用资源受限的进化算法来找到最优架构(即,子网络)。实验结果表明,继承自超网络的权值结构在KITTI上实现了4.67%的F1-all错误,RAFT基线减少了8.4%,超过了最先进的手工模型GMA和AGFlow,同时降低了模型复杂度和延迟。源代码和训练模型将在https://github.com/VDIGPKU/FlowNAS上发布。

英文摘要:

Existing optical flow estimators usually employ the network architectures typically designed for image classification as the encoder to extract per-pixel features. However, due to the natural difference between the tasks, the architectures designed for image classification may be sub-optimal for flow estimation. To address this issue, we propose a neural architecture search method named FlowNAS to automatically find the better encoder architecture for flow estimation task. We first design a suitable search space including various convolutional operators and construct a weight-sharing super-network for efficiently evaluating the candidate architectures. Then, for better training the super-network, we propose Feature Alignment Distillation, which utilizes a well-trained flow estimator to guide the training of super-network. Finally, a resource-constrained evolutionary algorithm is exploited to find an optimal architecture (i.e., sub-network). Experimental results show that the discovered architecture with the weights inherited from the super-network achieves 4.67\% F1-all error on KITTI, an 8.4\% reduction of RAFT baseline, surpassing state-of-the-art handcrafted models GMA and AGFlow, while reducing the model complexity and latency. The source code and trained models will be released in https://github.com/VDIGPKU/FlowNAS.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑