SiamL-MLP: Siamese-like Network for Object Tracking Based on MLP-Mixer
Shipeng Sun, Shiyong Lan, Weikang Huang, Piaoyang Li · 2022
Visual object tracking has always been a research hotspot in the field of computer vision. Recently, some excellent techniques in video object detection can be used to improve the performance of visual object trackers. Therefore, we propose a new Siamese-like network tracker by leveraging the superior representation ability of the recent visual object detection algorithm MLP-Mixer. Specifically, the MLP-Mixer based backbone network is used for feature extraction, which can enhance the feature representation. Furthermore, to connect the target template and the search region, we adopt cross-attention instead of conventional correlation operation to avoid getting stuck in local optimization when predicting target location. Finally, in order to demonstrate the effectiveness of the proposed method, we conduct extensive experiments on four tracking benchmarks, including OTB2015, VOT2016, VOT201S and GOT-10k datasets. Experimental results show that our proposed SiamL-MLP achieves competitive performance on the OTB2015, VOT2016, VOT201S and GOT-10k datasets. Our tracker runs at approximatively 40 FPS on GPU. Code will be available at https://github.com/SilvesterSun/SiamL-MLP.