Spatially Adaptive Computation Time for Residual Networks
This work addresses computational efficiency for computer vision tasks, offering a problem-agnostic solution applicable to image classification, object detection, and segmentation, though it is incremental as it builds on existing Residual Networks.
The paper tackles the problem of computational inefficiency in deep learning by proposing a Residual Network architecture that dynamically adjusts the number of layers per image region, improving efficiency on ImageNet classification and COCO object detection datasets with concrete gains.
This paper proposes a deep learning architecture based on Residual Network that dynamically adjusts the number of executed layers for the regions of the image. This architecture is end-to-end trainable, deterministic and problem-agnostic. It is therefore applicable without any modifications to a wide range of computer vision problems such as image classification, object detection and image segmentation. We present experimental results showing that this model improves the computational efficiency of Residual Networks on the challenging ImageNet classification and COCO object detection datasets. Additionally, we evaluate the computation time maps on the visual saliency dataset cat2000 and find that they correlate surprisingly well with human eye fixation positions.