Research map / Detection and segmentation / CNN features
YOLObile: Real-Time Object Detection on Mobile Devices via Compression-Compilation Co-Design
Yuxuan Cai, Hongjia Li, Geng Yuan, Wei Niu, Yanyu Li, Xulong Tang, Bin Ren, Yanzhi Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add GPU-CPU collaborative execution | YOLObile · Samsung Galaxy S20 | FPS: 17 -> 19.1 FPS | Improved |
| Apply block-punched pruning to YOLOv4 | YOLOv4 · MS COCO | mAP: 57.3 -> 49 | Hurt |
| Apply block-punched pruning to YOLOv4 | YOLOv4 · Samsung Galaxy S20 | FPS: 3.5 -> 17 | Improved |
| Replace block-punched pruning with unstructured pruning (ablation) | YOLOv4 at 8.09× compression · MS COCO | mAP: 51.4 -> 53.9 | Improved |
| Replace block-punched pruning with unstructured pruning (ablation) | YOLOv4 at 8.09× compression · Samsung Galaxy S20 | FPS: 11.5 -> 6.4 | Hurt |
| Replace block-punched pruning with structured filter pruning (ablation) | YOLOv4 at 8.09× compression · MS COCO | mAP: 51.4 -> 38.6 | Hurt |
| Replace block-punched pruning with structured filter pruning (ablation) | YOLOv4 at 8.09× compression · Samsung Galaxy S20 | FPS: 11.5 -> 13 | Improved |
| Increase block size (ablation) | Block-punched-pruned YOLOv4 · MS COCO and Samsung Galaxy S20 | mAP and FPS | Mixed |
| Use 8×4 blocks instead of other tested block sizes (ablation) | Block-punched-pruned YOLOv4 · MS COCO and Samsung Galaxy S20 | mAP and FPS | Mixed |
| Use pattern-based pruning instead of block-punched pruning (ablation) | YOLOv4 at compression rates below 3× · MS COCO | mAP | Improved |
| Use pattern-based pruning instead of block-punched pruning (ablation) | YOLOv4 · MS COCO and Samsung Galaxy S20 | mAP and FPS | Mixed |
| Prune 3×3 layers more heavily than 1×1 layers (ablation) | YOLOv4 at 6.2G FLOPs · MS COCO | mAP: 50.5 -> 51.4 | Improved |
Closest papers by meaning
- YOLO-ULM: Ultra-Lightweight Models for Real-Time Object Detection (CVPR 2026 · similarity 0.87)
- YOLOv10: Real-Time End-to-End Object Detection (NeurIPS 2024 · similarity 0.86)
- Mamba YOLO: A Simple Baseline for Object Detection with State Space Model (AAAI 2025 · similarity 0.86)
- DETRs Beat YOLOs on Real-time Object Detection (CVPR 2024 · similarity 0.83)
- SMamba: Sparse Mamba for Event-based Object Detection (AAAI 2025 · similarity 0.83)
- YOLO-World: Real-Time Open-Vocabulary Object Detection (CVPR 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
