Dense visual prediction tasks, including object detection and segmentation, inherently require precise and discriminative positional information to delineate object boundaries and pixel regions. Recent DETR-based frameworks advance dense prediction tasks through iterative attention applied to content queries, with sampled proposals as position references. However, this paradigm suffers from the misaligned sampling distribution and insufficient interaction between the content and position features, thereby limiting the encoding effectiveness. To overcome these limitations, we investigate the encoding paradigm for content-position harmonization and propose an effective predictor for dense visual tasks, termed DAPE (DETR with hArmonized content-Position Encoding). DAPE introduces explicit position encoding to facilitate content enhancement while maintaining low memory overhead. To achieves this process, DAPE comprises a Shifted Query Sampler (SQS) that enforces strict alignment between the distributions of content and position queries, and a 2D Low-Rank Position Encoder (LRPE) that progressively modulates attention maps based on the aligned representations. DAPE provides a unified solution for various dense prediction tasks. Extensive experiments on object detection, instance segmentation, and few-shot detection benchmarks demonstrate that DAPE achieves state-of-the-art performance while reducing memory consumption.

Surface Defect Detection (SDD) aims to accurately localize defects based on predefined category labels in industrial manufacturing. Different from generic object detection, the industrial environment introduces significant challenges due to interference and unrelated background textures, leading to increased confusion between defect and non-defect features. In this work, we identify and analyze the structural characteristics and relations inherent in defect features. This analysis enables effectively distinguishing defects from non-defect areas, thereby enhancing the discriminative power for surface defect detection. Based on this insight, we propose a novel surface defect detection framework, named UnfoldDet. This framework focuses on separating defect and non-defect features and reasoning about the relations among defects. Specifically, we formulate the feature separation as an optimization problem with structural constraints. By expressing its iterations as network stages, we introduce an unfolding fusion module (UFM) to progressively separate and fuse multi-scale features. At the instance level, we propose a hierarchical relation encoder (HRE) to capture the inherent relations among defect instances. Through reasoning on positional and categorical relations, only highly related defect features are enhanced, while unrelated non-defect features are suppressed. Through extensive quantitative and qualitative experiments, as well as ablation studies on real-world datasets including ESD, CSD, and NEU-DET, we demonstrate the effectiveness of the proposed UnfoldDet in terms of both performance and computational efficiency. The code is available at https://github.com/xiuqhou/UnfoldDet.

DETR-like methods have significantly increased detection performance in an end-to-end manner. The main-stream two-stage frameworks of them perform dense self-attention and select a fraction of queries for sparse cross-attention, which is proven effective for improving performance but also introduces a heavy computational burden and high dependence on stable query selection. This paper demonstrates that suboptimal two-stage selection strategies result in scale bias and redundancy due to the mismatch between selected queries and objects in two-stage initial-ization. To address these issues, we propose hierarchical salience filtering refinement, which performs transformer encoding only on filtered discriminative queries, for a bet-ter trade-off between computational efficiency and precision. The filtering process overcomes scale bias through a novel scale-independent salience supervision. To com-pensate for the semantic misalignment among queries, we introduce elaborate query refinement modules for stable two-stage initialization. Based on above improvements, the proposed Salience DETR achieves significant improvements of +4.0% AP, +0.2% AP, +4.4% AP on three challenging task-specific detection datasets, as well as 49.2% AP on COCO 2017 with less FLOPs. The code is available at https://github.com/xiuqhou/Salience-DETR.