YOLO-NAS-Bench: A Surrogate Benchmark with Self-Evolving Predictors for YOLO Architecture Search

2026-03-10

AI Summary

This work addresses the limitations of neural architecture search (NAS) for object detection, particularly in the context of YOLO models, which suffer from high training costs and the absence of dedicated benchmarks. To this end, we propose YOLO-NAS-Bench, the first proxy benchmark tailored for YOLO detectors, defining a comprehensive search space encompassing both backbone and neck components and training 1,000 sampled architectures on COCO-mini. We develop a LightGBM-based surrogate predictor enhanced with a self-evolution mechanism that actively guides the discovery of high-performing architectures and dynamically aligns the training distribution with the high-performance frontier. By integrating hierarchical sampling with evolutionary search, our surrogate achieves substantially improved predictive performance—R² increases from 0.770 to 0.815 and Sparse Kendall Tau from 0.694 to 0.752—and the discovered architectures outperform all official baselines from YOLOv8 to YOLO12 on COCO-mini.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Object Detection & Categorization

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Economics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd work

Abstract

Neural Architecture Search (NAS) for object detection is severely bottlenecked by high evaluation cost, as fully training each candidate YOLO architecture on COCO demands days of GPU time. Meanwhile, existing NAS benchmarks largely target image classification, leaving the detection community without a comparable benchmark for NAS evaluation. To address this gap, we introduce YOLO-NAS-Bench, the first surrogate benchmark tailored to YOLO-style detectors. YOLO-NAS-Bench defines a search space spanning channel width, block depth, and operator type across both backbone and neck, covering the core modules of YOLOv8 through YOLO12. We sample 1,000 architectures via random, stratified, and Latin Hypercube strategies, train them on COCO-mini, and build a LightGBM surrogate predictor. To sharpen the predictor in the high-performance regime most relevant to NAS, we propose a Self-Evolving Mechanism that progressively aligns the predictor's training distribution with the high-performance frontier, by using the predictor itself to discover and evaluate informative architectures in each iteration. This method grows the pool to 1,500 architectures and raises the ensemble predictor's R2 from 0.770 to 0.815 and Sparse Kendall Tau from 0.694 to 0.752, demonstrating strong predictive accuracy and ranking consistency. Using the final predictor as the fitness function for evolutionary search, we discover architectures that surpass all official YOLOv8-YOLO12 baselines at comparable latency on COCO-mini, confirming the predictor's discriminative power for top-performing detection architectures.

Problem

• Neural Architecture Search
• Object Detection
• YOLO
• Evaluation Cost
• NAS Benchmark

Innovation

• Surrogate Benchmark
• Neural Architecture Search
• YOLO
• Self-Evolving Mechanism
• Object Detection