FaPN: Feature-aligned Pyramid Network for Dense Image Prediction
This addresses misclassifications on object boundaries in dense prediction tasks like segmentation, offering incremental improvements over existing methods.
The paper tackles the problem of feature misalignment in dense image prediction by proposing a feature alignment module and a feature selection module integrated into a pyramid network, resulting in improvements of 1.2-2.6 points in AP/mIoU over FPN and achieving state-of-the-art 56.7% mIoU on ADE20K.
Recent advancements in deep neural networks have made remarkable leap-forwards in dense image prediction. However, the issue of feature alignment remains as neglected by most existing approaches for simplicity. Direct pixel addition between upsampled and local features leads to feature maps with misaligned contexts that, in turn, translate to mis-classifications in prediction, especially on object boundaries. In this paper, we propose a feature alignment module that learns transformation offsets of pixels to contextually align upsampled higher-level features; and another feature selection module to emphasize the lower-level features with rich spatial details. We then integrate these two modules in a top-down pyramidal architecture and present the Feature-aligned Pyramid Network (FaPN). Extensive experimental evaluations on four dense prediction tasks and four datasets have demonstrated the efficacy of FaPN, yielding an overall improvement of 1.2 - 2.6 points in AP / mIoU over FPN when paired with Faster / Mask R-CNN. In particular, our FaPN achieves the state-of-the-art of 56.7% mIoU on ADE20K when integrated within Mask-Former. The code is available from https://github.com/EMI-Group/FaPN.