# Scale-Equalizing Pyramid Convolution for Object Detection

arXiv: https://arxiv.org/abs/2005.03101 · HTML: https://mlsanity.com/paper/2005.03101/

## Key facts

- Cited by 25 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 6 May 2020.

## Built on by (most cited first, 25 of 25)

- [Dynamic Head: Unifying Object Detection Heads with Attentions](https://mlsanity.com/paper/2106.08322/) (2021)
- [AdaMixer: A Fast-Converging Query-Based Object Detector](https://mlsanity.com/paper/2203.16507/) (2022)
- [Dense Distinct Query for End-to-End Object Detection](https://mlsanity.com/paper/2303.12776/) (2023)
- [QueryDet: Cascaded Sparse Query for Accelerating High-Resolution Small Object Detection](https://mlsanity.com/paper/2103.09136/) (2021)
- [DEVIANT: Depth EquiVarIAnt NeTwork for Monocular 3D Object Detection](https://mlsanity.com/paper/2207.10758/) (2022)
- [DISCO: accurate Discrete Scale Convolutions](https://mlsanity.com/paper/2106.02733/) (2021)
- [Resolution learning in deep convolutional networks using scale-space theory](https://mlsanity.com/paper/2106.03412/) (2021)
- [Object Detection with Transformers: A Review](https://mlsanity.com/paper/2306.04670/) (2023)
- [OPANAS: One-Shot Path Aggregation Network Architecture Search for Object Detection](https://mlsanity.com/paper/2103.04507/) (2021)
- [Efficient Self-Ensemble for Semantic Segmentation](https://mlsanity.com/paper/2111.13280/) (2021)
- [Learning to Aggregate Multi-Scale Context for Instance Segmentation in Remote Sensing Images](https://mlsanity.com/paper/2111.11057/) (2021)
- [USB: Universal-Scale Object Detection Benchmark](https://mlsanity.com/paper/2103.14027/) (2021)
- [RCNet: Reverse Feature Pyramid and Cross-scale Shift Network for Object Detection](https://mlsanity.com/paper/2110.12130/) (2021)
- [Large-batch Optimization for Dense Visual Predictions](https://mlsanity.com/paper/2210.11078/) (2022)
- [DRPN: Making CNN Dynamically Handle Scale Variation](https://mlsanity.com/paper/2112.10963/) (2021)
- [Scale generalisation properties of extended scale-covariant and scale-invariant Gaussian derivative networks on image datasets with spatial scaling variations](https://mlsanity.com/paper/2409.11140/) (2024)
- [ssFPN: Scale Sequence (S^2) Feature Based-Feature Pyramid Network for Object Detection](https://mlsanity.com/paper/2208.11533/) (2022)
- [LAPFormer: A Light and Accurate Polyp Segmentation Transformer](https://mlsanity.com/paper/2210.04393/) (2022)
- [MRC-DETR: An Adaptive Multi-Residual Coupled Transformer for Bare Board PCB Defect Detection](https://mlsanity.com/paper/2507.03386/) (2025)
- [AGSFCOS: Based on attention mechanism and Scale-Equalizing pyramid network of object detection](https://mlsanity.com/paper/2105.09596/) (2021)
- [MPSN: Motion-aware Pseudo Siamese Network for Indoor Video Head Detection in Buildings](https://mlsanity.com/paper/2110.03302/) (2021)
- [HR-RCNN: Hierarchical Relational Reasoning for Object Detection](https://mlsanity.com/paper/2110.13892/) (2021)
- [Delving into the Scale Variance Problem in Object Detection](https://mlsanity.com/paper/2206.08227/) (2022)
- [Scale-invariant Gaussian derivative residual networks](https://mlsanity.com/paper/2603.02843/) (2026)
- [A3-FPN: Asymptotic Content-Aware Pyramid Attention Network for Dense Visual Prediction](https://mlsanity.com/paper/2604.10210/) (2026)

## Abstract (arXiv)

Feature pyramid has been an efficient method to extract features at different scales. Development over this method mainly focuses on aggregating contextual information at different levels while seldom touching the inter-level correlation in the feature pyramid. Early computer vision methods extracted scale-invariant features by locating the feature extrema in both spatial and scale dimension. Inspired by this, a convolution across the pyramid level is proposed in this study, which is termed pyramid convolution and is a modified 3-D convolution. Stacked pyramid convolutions directly extract 3-D (scale and spatial) features and outperforms other meticulously designed feature fusion modules. Based on the viewpoint of 3-D convolution, an integrated batch normalization that collects statistics from the whole feature pyramid is naturally inserted after the pyramid convolution. Furthermore, we also show that the naive pyramid convolution, together with the design of RetinaNet head, actually best applies for extracting features from a Gaussian pyramid, whose properties can hardly be satisfied by a feature pyramid. In order to alleviate this discrepancy, we build a scale-equalizing pyramid convolution (SEPC) that aligns the shared pyramid convolution kernel only at high-level feature maps. Being computationally efficient and compatible with the head design of most single-stage object detectors, the SEPC module brings significant performance improvement (>4AP increase on MS-COCO2017 dataset) in state-of-the-art one-stage object detectors, and a light version of SEPC also has ~3.5AP gain with only around 7% inference time increase. The pyramid convolution also functions well as a stand-alone module in two-stage object detectors and is able to improve the performance by ~2AP. The source code can be found at https://github.com/jshilong/SEPC.