@article{ART003317584},
author={Min-Ho Kim and Kyu-Cheol Cho},
title={Training-Free Feature-Level Interpolation with Sparse-Backbone Scheduling for YOLO Detection in Low-FPS Video},
journal={Journal of The Korea Society of Computer and Information},
issn={1598-849X},
year={2026},
volume={31},
number={3},
pages={49-59}
TY - JOUR
AU - Min-Ho Kim
AU - Kyu-Cheol Cho
TI - Training-Free Feature-Level Interpolation with Sparse-Backbone Scheduling for YOLO Detection in Low-FPS Video
JO - Journal of The Korea Society of Computer and Information
PY - 2026
VL - 31
IS - 3
PB - The Korean Society Of Computer And Information
SP - 49
EP - 59
SN - 1598-849X
AB - In this paper, we propose a training-free method that mitigates degradation in accuracy and temporal stability while improving system efficiency for YOLO-based detectors in low-frame-rate (Low-FPS) environments. The proposed approach preserves the YOLO backbone and neck, caches feature maps (P3 –P5) right before the Detection Head at Anchor frames, and operates in a Sparse-Backbone manner by performing head-only inference on linearly interpolated features from neighboring Anchors for non-anchor frames. Under a system-level end-to-end(E2E) protocol (5 independent runs, mean ± std), it reduces backbone invocations by about 50% while retaining 98.48% of the baseline mean Average Precision at IoU 0.5 (mAP@50), decreasing processing latency (Lat_proc) by 26.60% and improving processing throughput (FPS_proc) by 36.11%. These results confirm that the proposed method can serve as a practical training-free inference module that preserves detection quality and temporal stability while improving system efficiency in Low-FPS and resource-constrained deployments.
KW - Feature-Level Interpolation;Low-FPS Video;YOLO;Training-Free Module;Object Detection
DO -
UR -
ER -
Min-Ho Kim and Kyu-Cheol Cho. (2026). Training-Free Feature-Level Interpolation with Sparse-Backbone Scheduling for YOLO Detection in Low-FPS Video. Journal of The Korea Society of Computer and Information, 31(3), 49-59.
Min-Ho Kim and Kyu-Cheol Cho. 2026, "Training-Free Feature-Level Interpolation with Sparse-Backbone Scheduling for YOLO Detection in Low-FPS Video", Journal of The Korea Society of Computer and Information, vol.31, no.3 pp.49-59.
Min-Ho Kim, Kyu-Cheol Cho "Training-Free Feature-Level Interpolation with Sparse-Backbone Scheduling for YOLO Detection in Low-FPS Video" Journal of The Korea Society of Computer and Information 31.3 pp.49-59 (2026) : 49.
Min-Ho Kim, Kyu-Cheol Cho. Training-Free Feature-Level Interpolation with Sparse-Backbone Scheduling for YOLO Detection in Low-FPS Video. 2026; 31(3), 49-59.
Min-Ho Kim and Kyu-Cheol Cho. "Training-Free Feature-Level Interpolation with Sparse-Backbone Scheduling for YOLO Detection in Low-FPS Video" Journal of The Korea Society of Computer and Information 31, no.3 (2026) : 49-59.
Min-Ho Kim; Kyu-Cheol Cho. Training-Free Feature-Level Interpolation with Sparse-Backbone Scheduling for YOLO Detection in Low-FPS Video. Journal of The Korea Society of Computer and Information, 31(3), 49-59.
Min-Ho Kim; Kyu-Cheol Cho. Training-Free Feature-Level Interpolation with Sparse-Backbone Scheduling for YOLO Detection in Low-FPS Video. Journal of The Korea Society of Computer and Information. 2026; 31(3) 49-59.
Min-Ho Kim, Kyu-Cheol Cho. Training-Free Feature-Level Interpolation with Sparse-Backbone Scheduling for YOLO Detection in Low-FPS Video. 2026; 31(3), 49-59.
Min-Ho Kim and Kyu-Cheol Cho. "Training-Free Feature-Level Interpolation with Sparse-Backbone Scheduling for YOLO Detection in Low-FPS Video" Journal of The Korea Society of Computer and Information 31, no.3 (2026) : 49-59.
