Research map / Detection and segmentation / CNN features
iNAS: Integral NAS for Device-Aware Salient Object Detection
Yu-Chao Gu, Shang-Hua Gao, Xu-Sheng Cao, Peng Du, Shao-Ping Lu, Ming-Ming Cheng
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 searchable backbone (ablation) | fixed backbone + fixed head · CPU latency, ECSSD | latency (ms): 45.17 -> 41.20 | Improved |
| add searchable saliency head (ablation) | fixed backbone + fixed head · CPU latency, ECSSD | latency (ms): 45.17 -> 36.20 | Improved |
| search backbone and head jointly (integral space) (ablation) | fixed backbone + fixed head · CPU latency, ECSSD | latency (ms): 45.17 -> 33.06 | Improved |
| search backbone and head jointly (integral space) (ablation) | fastest architecture · ECSSD | maxF: 0.941 -> 0.944 | Improved |
| search backbone and head jointly (integral space) (ablation) | best-performance architecture · ECSSD | maxF: 0.941 -> 0.947 | Improved |
| search only the backbone (ablation) | fixed backbone + fixed head · ECSSD | maxF: 0.941 -> 0.941 | No clear effect |
| use latency-group sampling instead of uniform sampling in evolution search (ablation) | evolution search · search space latency range | coverage of latency range: 50.2% -> 99% | Improved |
| use device-aware search on CPU instead of GPU-searched model (ablation) | iNAS(GPU)-L · CPU latency | latency (ms): 48.90 -> 42.99 | Improved |
| use device-aware search on mobile phone instead of GPU-searched model (ablation) | iNAS(GPU)-L · Mobile latency | latency (ms): 397.17 -> 339.61 | Improved |
| use device-aware search on embedded device instead of GPU-searched model (ablation) | iNAS(GPU)-L · Embedded latency | latency (ms): 71.70 -> 63.39 | Improved |
| use searchable multi-scale unit instead of inverted bottleneck (ablation) | search space · latency-maxF Pareto frontier (Fig. 10) | maxF vs latency | Improved |
| increase backbone complexity (ablation) | iNAS sampled models · 2000 sampled models (Fig. 11) | performance | Improved |
Closest papers by meaning
- NAS-FCOS: Fast Neural Architecture Search for Object Detection (CVPR 2020 · similarity 0.86)
- MnasFPN: Learning Latency-Aware Pyramid Architecture for Object Detection on Mobile Devices (CVPR 2020 · similarity 0.85)
- NAS-FCOS: Efficient Search for Object Detection Architectures (IJCV 2021 · similarity 0.85)
- Salient Object Detection in the Deep Learning Era: An In-Depth Survey (TPAMI 2021 · similarity 0.84)
- Exploring Salient Object Detection with Adder Neural Networks (AAAI 2025 · similarity 0.83)
- Samba: A Unified Mamba-based Framework for General Salient Object Detection (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.
