LeYOLO: Lightweight, Scalable and Efficient CNN Architecture for Object Detection

Lilian Hollard, Lucas Mohimont, Nathalie Gaveau, Luiz Angelo Steffenel

Keywords: Object detection, CNN, parameter efficiency, Architecture optimization

Abstract:

Computational efficiency in deep neural networks is critical for object detection, especially as newer models prioritize speed over efficient computation (Parameters and FLOP). This trend is evident in the latest YOLO architectures, which focus more on speed at the expense of lightweight design. This evolution has somewhat left lightweight architecture design behind for object detection applications.
Unlike speed-oriented object detectors in the literature, SSDLite and low-parameters/FLOP-oriented classifier combinations are the only proposed solutions, leaving a gap between YOLO-like architectures and lightweight object detectors. In this paper, we pose the question: $\textit{Can an architecture optimized for parameters and FLOPs achieve precision comparable to mainstream YOLO models?}$ To explore this, we introduce LeYOLO, an efficient object detection model, and propose several optimizations to enhance the computational efficiency of YOLO-based models. This approach bridges the gap between SSDLite-based object detectors and YOLO models, achieving high precision in a model as lightweight as MobileNets.
Our novel model family achieves a FLOP-to-accuracy ratio previously unattained, offering scalability that spans from ultra-low neural network configurations $( \(<\) 1 GFLOP)$ to efficient yet demanding object detection setups $( \(>\) 4 GFLOPs)$ with 25.2, 31.3, 35.2, 38.2, 39.3 and 41 mAP for 0.66, 1.47, 2.53, 4.51, 5.8 and 8.4 FLOP(G).

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9209