ContextFlow-YOLO: Context Preservation and Cross-Scale Redistribution for Small-Object Detection in Complex Environments
- 1Center for Artificial Intelligence Technology, Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, Bangi 43600, Malaysia
- 2School of Computer Science and Artificial Intelligence, Chaohu University, Hefei 238024, China
Abstract
Small objects in underwater and traffic-sign images often have weak boundaries and low contrast. Downsampling and multi-scale fusion can further weaken the features needed to detect them. We propose ContextFlow-YOLO, a YOLOv8n-based detector designed to preserve context during feature extraction and reuse it across scales. Its context-preserving backbone combines local features with surrounding and global information during downsampling. The global-guided scale-sequence neck distributes this context to features at different levels. A nonlinear context-attention fusion unit then refines the responses used for prediction. ContextFlow-YOLO obtains [email protected] scores of 0.855, 0.838, and 0.826 on URPC2019, URPC2020, and TT100K, respectively. On TT100K, the model uses 6.11 M parameters and runs at 134 FPS. Heatmap and receptive-field visualizations show more concentrated target responses and broader feature coverage than YOLOv8n.
Related
- Detection Augmented Learning In Non-Stationary Bandit Environments
- Context-Aware Detection of Content Evidencing Illegal or Predatory Fintech Lending Practices on Indonesian Digital Platforms Using Transformer-Based Language Models
- Context-Guided Traversal with Deep Reinforcement Learning for Suspicious Subgraph Detection in Cryptocurrency Transaction Graphs
- Delta Redistribution: a variation of State Redistribution for steady-state embedded boundary computations
- Adaptive Randomized Neural Networks for PDEs: Structure-Aware Feature Redistribution and Residual-Driven Block Enrichment
BibTeX
@article{jiao2026contextflowyolo,
title = {{ContextFlow-YOLO: Context Preservation and Cross-Scale Redistribution for Small-Object Detection in Complex Environments}},
author = {Yuqing Jiao and Siok Yee Tan and Fadilla ’Atyka Nor Rashid and Huifen Xing and Yong Zhang},
journal = {Mathematics},
year = {2026},
month = oct,
doi = {10.3390/math14203781},
url = {https://www.mdpi.com/2227-7390/14/20/3781},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - ContextFlow-YOLO: Context Preservation and Cross-Scale Redistribution for Small-Object Detection in Complex Environments
AU - Yuqing Jiao
AU - Siok Yee Tan
AU - Fadilla ’Atyka Nor Rashid
AU - Huifen Xing
AU - Yong Zhang
PY - 2026
DA - 2026/10/18
JO - Mathematics
DO - 10.3390/math14203781
UR - https://www.mdpi.com/2227-7390/14/20/3781
AB - Small objects in underwater and traffic-sign images often have weak boundaries and low contrast. Downsampling and multi-scale fusion can further weaken the features needed to detect them. We propose ContextFlow-YOLO, a YOLOv8n-based detector designed to preserve context during feature extraction and reuse it across scales. Its context-preserving backbone combines local features with surrounding and global information during downsampling. The global-guided scale-sequence neck distributes this context to features at different levels. A nonlinear context-attention fusion unit then refines the responses used for prediction. ContextFlow-YOLO obtains [email protected] scores of 0.855, 0.838, and 0.826 on URPC2019, URPC2020, and TT100K, respectively. On TT100K, the model uses 6.11 M parameters and runs at 134 FPS. Heatmap and receptive-field visualizations show more concentrated target responses and broader feature coverage than YOLOv8n.
SN - 2227-7390
ER -