Linjie Luo

Finding visual correspondence between local features is key to many computer vision problems. While defining features with larger contextual scales usually implies greater discriminativeness, it could also lead to less spatial accuracy of the features. We propose AutoScaler, a scale-attention network to explicitly optimize this trade-off in visual correspondence tasks. Our architecture consists of a weight-sharing feature network to compute multi-scale feature maps and an attention network to combine them optimally in the scale space. This allows our network to have adaptive sizes of equivalent receptive field over different scales of the input. The entire network can be trained end-to-end in a Siamese framework for visual correspondence tasks. Using the latest off-the-shelf architecture for the feature network, our method achieves competitive results compared to state-of-the-art methods on challenging optical flow and semantic matching bench- marks, including Sintel, KITTI and CUB-2011. We also show that our attention network alone can be applied to existing hand-crafted feature descriptors (e.g Daisy) and improve their performance on visual correspondence tasks. Finally, we illustrate how the scale- attention maps generated from the attention network are visually interpretable.