Depth-Adaptive Computational Policies for Efficient Visual Tracking
This work addresses computational inefficiency in visual tracking for real-time applications, offering a runtime trade-off between speed and accuracy, though it is incremental as it builds on existing Siamese network methods.
The paper tackles the problem of inefficient computation in video object tracking by proposing a depth-adaptive convolutional Siamese network that adjusts neural network depth per frame based on difficulty, achieving accuracy comparable to state-of-the-art on VOT2016 while reducing computational cost.
Current convolutional neural networks algorithms for video object tracking spend the same amount of computation for each object and video frame. However, it is harder to track an object in some frames than others, due to the varying amount of clutter, scene complexity, amount of motion, and object's distinctiveness against its background. We propose a depth-adaptive convolutional Siamese network that performs video tracking adaptively at multiple neural network depths. Parametric gating functions are trained to control the depth of the convolutional feature extractor by minimizing a joint loss of computational cost and tracking error. Our network achieves accuracy comparable to the state-of-the-art on the VOT2016 benchmark. Furthermore, our adaptive depth computation achieves higher accuracy for a given computational cost than traditional fixed-structure neural networks. The presented framework extends to other tasks that use convolutional neural networks and enables trading speed for accuracy at runtime.