Research map / Detection and segmentation / CNN features
AutoFocus: Efficient Multi-Scale Inference
Mahyar Najibi, Bharat Singh, Larry S. Davis
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 |
|---|---|---|---|
| apply AutoFocus FocusChips at finer scales instead of full 3-scale pyramid | SNIPER detector · COCO test-dev | AP: 47.9 -> 47.9 at 1175^2 pixels vs 1910^2 | No clear effect |
| use AutoFocus for faster inference | SNIPER detector · COCO test-dev | images per second: 2.5 -> 6.4 images per second | Improved |
| reduce pixels processed to 930^2 | AutoFocus · COCO test-dev | AP: 47.9 -> 47.2 | Hurt |
| reduce pixels processed to 860^2 | AutoFocus · COCO test-dev | AP: 47.9 -> 46.9 | Hurt |
| add scale 3 (highest resolution) to the lower two scales (ablation) | multi-scale SNIPER · COCO val-2017 | AP: 45.4 -> 47.5 | Improved |
| add FocusChip processing of one fifth of scale 3 to scales 1 and 2 (ablation) | multi-scale SNIPER · COCO val-2017 | AP: 45 -> 47.5 | Improved |
| reduce pixels at highest resolution by 11 times with FocusChips (ablation) | AutoFocus · COCO val-2017 | AP: 0.5% drop in mAP | Hurt |
| reduce overall pixels processed by 5 times (ablation) | AutoFocus · COCO val-2017 | AP: 1% drop in mAP | Hurt |
| reduce pixels processed by 2.8 times (ablation) | AutoFocus · COCO val-2017 | AP: 2.8 times fewer pixels without any loss of performance | No clear effect |
| use COCO-tuned hyper-parameters on PASCAL VOC (AutoFocus*) instead of SNIPER full processing | AutoFocus · PASCAL VOC 2007 test | AP70: 80.5 -> 79.5 | Hurt |
| use AutoFocus instead of Deformable ConvNet v2 at similar processed area | detector · PASCAL VOC 2007 test | AP70: 4.6% better AP at 0.7 IoU | Improved |
| reduce minimum chip size from 512 to 64 (ablation) | FocusChip generation · COCO | theoretical speedup: theoretical speedup of ~10 times but detection performance can drop | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SNIPER — SNIPER: Efficient Multi-Scale Training (NeurIPS 2018) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning to Autofocus (CVPR 2020 · similarity 0.88)
- Scale Normalized Image Pyramids with AutoFocus for Object Detection (TPAMI 2021 · similarity 0.88)
- Stabilizing and Accelerating Autofocus with Expert Trajectory Regularized Deep Reinforcement Learning (CVPR 2025 · similarity 0.84)
- DualFocus: Depth from Focus with Spatio-Focal Dual Variational Constraints (NeurIPS 2025 · similarity 0.84)
- Multi-Focus Image Fusion via Explicit Defocus Blur Modelling (AAAI 2025 · similarity 0.83)
- One-Step Event-Driven High-Speed Autofocus (CVPR 2025 · similarity 0.83)
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.
