Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Confident Learning for Object Detection under Model Constraints

Yingda Yu, Jiaqi Xuan, Shuhui Shi, Xuanyu Teng, Shuyang Xu, Guanchao Tong

TL;DR

This work tackles real-time weed detection on resource-limited edge devices by addressing data quality rather than enlarging models. It introduces Model-Driven Data Correction (MDDC), a data-centric, six-stage workflow that couples a YOLO-based detector with Cluster-Level Outlier Detection to automatically diagnose and correct annotation noise, using train–fix–retrain iterations. Across four diverse weed/crop datasets, MDDC delivers consistent $5$–$25\%$ gains in $\mathrm{mAP}_{@0.5}$ and improvements in $\mathrm{mAP}_{@0.5:0.95}$, precision, recall, and F1-score, outperforming several state-of-the-art noise-handling methods while preserving a fixed model capacity. The results demonstrate that systematic data quality optimization can yield substantial practical benefits for precision agriculture on edge devices, with a reproducible workflow and accessible code.

Abstract

Agricultural weed detection on edge devices is subject to strict constraints on model capacity, computational resources, and real-time inference latency, which prevent performance improvements through model scaling or ensembling. This paper proposes Model-Driven Data Correction (MDDC), a data-centric framework that enhances detection performance by iteratively diagnosing and correcting data quality deficiencies. An automated error analysis procedure categorizes detection failures into four types: false negatives, false positives, class confusion, and localization errors. These error patterns are systematically addressed through a structured train-fix-retrain pipeline with version-controlled data management. Experimental results on multiple weed detection datasets demonstrate consistent improvements of 5-25 percent in mAP at 0.5 using a fixed lightweight detector (YOLOv8n), indicating that systematic data quality optimization can effectively alleviate performance bottlenecks under fixed model capacity constraints.

Confident Learning for Object Detection under Model Constraints

TL;DR

This work tackles real-time weed detection on resource-limited edge devices by addressing data quality rather than enlarging models. It introduces Model-Driven Data Correction (MDDC), a data-centric, six-stage workflow that couples a YOLO-based detector with Cluster-Level Outlier Detection to automatically diagnose and correct annotation noise, using train–fix–retrain iterations. Across four diverse weed/crop datasets, MDDC delivers consistent – gains in and improvements in , precision, recall, and F1-score, outperforming several state-of-the-art noise-handling methods while preserving a fixed model capacity. The results demonstrate that systematic data quality optimization can yield substantial practical benefits for precision agriculture on edge devices, with a reproducible workflow and accessible code.

Abstract

Agricultural weed detection on edge devices is subject to strict constraints on model capacity, computational resources, and real-time inference latency, which prevent performance improvements through model scaling or ensembling. This paper proposes Model-Driven Data Correction (MDDC), a data-centric framework that enhances detection performance by iteratively diagnosing and correcting data quality deficiencies. An automated error analysis procedure categorizes detection failures into four types: false negatives, false positives, class confusion, and localization errors. These error patterns are systematically addressed through a structured train-fix-retrain pipeline with version-controlled data management. Experimental results on multiple weed detection datasets demonstrate consistent improvements of 5-25 percent in mAP at 0.5 using a fixed lightweight detector (YOLOv8n), indicating that systematic data quality optimization can effectively alleviate performance bottlenecks under fixed model capacity constraints.
Paper Structure (21 sections, 6 equations, 3 figures, 4 tables, 2 algorithms)

This paper contains 21 sections, 6 equations, 3 figures, 4 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Rule-based and Human-in-the-loop Data Cleaning
  4. Noise-aware Sample Filtering
  5. Loss Reweighting and Sample-weighting
  6. Automatic Annotation Correction
  7. Method
  8. Dataset Preparation
  9. Baseline Training
  10. Spatial Clustering and Reduction
  11. Noise Analysis
  12. Label Correction
  13. Dataset Retraining
  14. Experiments and Results
  15. Experimental Setup
  16. ...and 6 more sections

Figures (3)

  • Figure 1: A proposed use-case of our framework
  • Figure 2: The complete workflow consists of Six sequential stages, including dataset preparation, baseline training, spatial clustering and reduction, noise analysis, Label correction, dataset retraining.
  • Figure 3: Examples of typical annotation cleaning results