Non-Maximum Suppression using Numpy

2026-05-18 · Source: Machine Learning on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Intermediate, short

Summary

Non-Maximum Suppression (NMS) is a critical algorithm used in object detection pipelines, particularly for models like YOLO, to eliminate redundant bounding box predictions around a single object. While high-level libraries offer built-in NMS functions, this article details a highly efficient, vectorized implementation using pure NumPy. The core NMS logic involves sorting bounding boxes by confidence scores, iteratively selecting the highest-scoring box, adding it to a "keep" list, and then suppressing all remaining boxes that have an Intersection over Union (IoU) with the selected box above a predefined threshold. The provided Python code demonstrates a `bbox_iou` function and the `nms` function, leveraging NumPy's broadcasting capabilities to perform simultaneous comparisons and optimize performance, avoiding computationally expensive nested `for` loops.

Key takeaway

For Machine Learning Engineers optimizing real-time object detection models, understanding and implementing vectorized NMS in NumPy is crucial. This approach enhances algorithmic efficiency by leveraging array broadcasting, which is vital for on-device inference and can significantly improve performance compared to naive loop-based methods. Mastering these mechanics provides a strong foundation for advanced optimizations in frameworks like PyTorch or TensorRT.

Key insights

Vectorized NumPy implementation of Non-Maximum Suppression efficiently cleans redundant object detection bounding boxes.

Principles

• Greedy selection optimizes NMS.
• Vectorization improves algorithmic efficiency.
• IoU is central to bounding box overlap.

Method

The NMS algorithm sorts bounding boxes by confidence, iteratively selects the highest, calculates IoU with remaining boxes, and suppresses those exceeding a threshold, repeating until no boxes remain.

In practice

• Use `np.maximum` and `np.minimum` for IoU.
• Apply boolean masks for fast indexing.
• Implement NMS for object detection cleanup.

Topics

• Non-Maximum Suppression
• Object Detection
• NumPy Vectorization
• Intersection over Union
• Bounding Box Prediction

Best for: Machine Learning Engineer, Computer Vision Engineer, AI Engineer

Related on AIssential

• Learned Non-Maximum Suppression for 3D Object Detection — Learned filtering modules, D2D-Rescore and GossipNet3D, enhance 3D object detection by replacing heuristic NMS with detection-to-detection relation leveraging.
• MoEIoU: Rethinking Bounding-Box Regression as a Mixture of Experts — MoEIoU adaptively prioritizes localization error components for improved bounding-box regression, leading to faster convergence and higher accuracy.
• 3 NumPy Tricks for Numerical Performance — Optimize NumPy performance by utilizing vectorization, in-place operations, and memory views to avoid Python overhead and unnecessary memory allocations.
• Edge-Constrained UAV Small-Object Detection with P2 Enhancement and Quantum-Inspired Lightweight Structure Search — Combining a P2 high-resolution branch with QIEA structure search enhances UAV small-object detection under edge constraints.
• A Probabilistic Framework for Improving Dense Object Detection in Underwater Image Data via Annealing-Based Data Augmentation — A pseudo-simulated annealing data augmentation method improves underwater object detection in crowded, natural scenes.
• Demystifying RF-DETR [ICLR 2026]: A Real-Time Transformer Pushing the Limits of Object Detection — RF-DETR uses weight-sharing NAS and internet-scale pre-training to achieve state-of-the-art real-time object detection and segmentation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning on Medium.