PGNeXt: High-Resolution Salient Object Detection via Pyramid Grafting Network

TL;DR

This work tackles the challenging problem of high-resolution salient object detection by introducing PGNeXt, a pyramid grafting network that unifies CNN and Transformer features through hierarchical staggered grafting and a window-based Cross-Model Grafting Module (wCMGM). Central to the approach are the Attention Guided Loss (AGL) and the newly released Ultra High-Resolution Saliency Detection Dataset (UHRSD) comprising 5,920 pixellabeled 4K–8K images, enabling robust learning of fine-grained details. Empirically, PGNeXt achieves state-of-the-art performance on three HR-SOD benchmarks, with especially notable improvements in boundary quality (mBA), and demonstrates strong generalization to camouflaged object detection without architectural changes. The combination of a high-resolution dataset and the grafting framework offers practical benefits for real-world HR-SOD applications where detail preservation and accurate edge delineation are critical.

Abstract

We present an advanced study on more challenging high-resolution salient object detection (HRSOD) from both dataset and network framework perspectives. To compensate for the lack of HRSOD dataset, we thoughtfully collect a large-scale high resolution salient object detection dataset, called UHRSD, containing 5,920 images from real-world complex scenarios at 4K-8K resolutions. All the images are finely annotated in pixel-level, far exceeding previous low-resolution SOD datasets. Aiming at overcoming the contradiction between the sampling depth and the receptive field size in the past methods, we propose a novel one-stage framework for HR-SOD task using pyramid grafting mechanism. In general, transformer-based and CNN-based backbones are adopted to extract features from different resolution images independently and then these features are grafted from transformer branch to CNN branch. An attention-based Cross-Model Grafting Module (CMGM) is proposed to enable CNN branch to combine broken detailed information more holistically, guided by different source feature during decoding process. Moreover, we design an Attention Guided Loss (AGL) to explicitly supervise the attention matrix generated by CMGM to help the network better interact with the attention from different branches. Comprehensive experiments on UHRSD and widely-used SOD datasets demonstrate that our method can simultaneously locate salient object and preserve rich details, outperforming state-of-the-art methods. To verify the generalization ability of the proposed framework, we apply it to the camouflaged object detection (COD) task. Notably, our method performs superior to most state-of-the-art COD methods without bells and whistles.

Figures (11)

• Figure 1: Comparison of the results of the different methods. (a) Input image. (b) Ground truth maks. (c) Down-sample then input to CNN-based PFSNet.(d) Directly input to ResNet-18 based FPN. (e) Down-sample then input to Swin-based FPN. (f) Ours.
• Figure 2: Comparison of different methods for HR inputs. (a) Down-sampling strategy. (b) Two-stage coarse to fine framework. (c) Recurrent framework (d) Pyramid grafting framework.
• Figure 3: Categories of our UHRSD dataset. We illustrate 7 primary categories. For items within primary categories that are hard to classify, we categorize them as "non-specific". The columns' heights in this chart approximate the distribution of quantities across categories.
• Figure 4: Examples and corresponding annotations in UHRSD. UHRSD contains rich salinent objects in terms of classes and attributes.
• Figure 5: Comparison of annotation quality among UHRSD and other SOD datasets. From left to right: 2 sample images from UHRSD; comparison of annotation quality between UHRSD and HRSOD; comparison of annotation quality between UHRSD and DUTS. Best viewed by zoom-in.