Fast Video Object Segmentation using the Global Context Module
This work addresses the need for efficient and accurate video object segmentation, which is incremental as it builds on prior methods by improving memory and computation efficiency.
The paper tackles the problem of semi-supervised video object segmentation by developing a real-time algorithm that matches the accuracy of slower online-learning models and the speed of faster but less accurate template-matching methods, achieving top performance on standard benchmarks.
We developed a real-time, high-quality semi-supervised video object segmentation algorithm. Its accuracy is on par with the most accurate, time-consuming online-learning model, while its speed is similar to the fastest template-matching method with sub-optimal accuracy. The core component of the model is a novel global context module that effectively summarizes and propagates information through the entire video. Compared to previous approaches that only use one frame or a few frames to guide the segmentation of the current frame, the global context module uses all past frames. Unlike the previous state-of-the-art space-time memory network that caches a memory at each spatio-temporal position, the global context module uses a fixed-size feature representation. Therefore, it uses constant memory regardless of the video length and costs substantially less memory and computation. With the novel module, our model achieves top performance on standard benchmarks at a real-time speed.