Laplacian Pyramid Reconstruction and Refinement for Semantic Segmentation
It addresses pixel-accurate semantic segmentation for computer vision applications, offering a novel method without complex inference or detection.
The paper tackled the problem of CNN spatial pooling reducing resolution for dense pixel labeling by showing that feature maps contain sub-pixel localization information and proposing a Laplacian pyramid reconstruction architecture with skip connections and gating to refine boundaries. This approach achieved state-of-the-art results on PASCAL VOC and Cityscapes benchmarks.
CNN architectures have terrific recognition performance but rely on spatial pooling which makes it difficult to adapt them to tasks that require dense, pixel-accurate labeling. This paper makes two contributions: (1) We demonstrate that while the apparent spatial resolution of convolutional feature maps is low, the high-dimensional feature representation contains significant sub-pixel localization information. (2) We describe a multi-resolution reconstruction architecture based on a Laplacian pyramid that uses skip connections from higher resolution feature maps and multiplicative gating to successively refine segment boundaries reconstructed from lower-resolution maps. This approach yields state-of-the-art semantic segmentation results on the PASCAL VOC and Cityscapes segmentation benchmarks without resorting to more complex random-field inference or instance detection driven architectures.