Research map / Detection and segmentation / Open-set and few-shot detection
Deep Multicameral Decoding for Localizing Unoccluded Object Instances from a Single RGB Image
Matthieu Grard, Emmanuel Dellandréa, Liming Chen
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 |
|---|---|---|---|
| use multicameral decoding instead of aggregated dilated convolutions | RED network · Mikado | AP: 20.6 points | Improved |
| use multicameral decoding instead of coordinate-aware convolutions | RED network · Mikado | AP: 7.8 points | Improved |
| use multicameral decoding instead of DenseNet encoder blocks | RED network · Mikado | AP: 5.1 points | Improved |
| add a third cascaded decoder (MC3) (ablation) | MC2 · Mikado | AP: +1.8 points | Improved |
| add a fourth cascaded decoder (MC4) (ablation) | MC2 · Mikado | AP: +3 points | Improved |
| add ordinal intermediate supervisions (ablation) | MC4 · Mikado | AP: +4 points | Improved |
| append a refinement encoder-decoder unit and supervisions (MC6) (ablation) | MC2 · Mikado | AP: +9.3 points | Improved |
| add atrous convolutions late in decoding (first decoder for unoccluded instances) (ablation) | MC6-dagger · Mikado | AP: +1.2 points | Improved |
| use bicameral ordinal decoders instead of DOOBNet/MTAN-like designs (ablation) | MC3-dagger · Mikado occlusion boundaries | AP60 occlusion: +1.7 points | Improved |
| use bicameral ordinal decoders instead of DOOBNet/MTAN-like designs (ablation) | MC3-dagger · PIOD occlusion boundaries | AP60 occlusion: +1 point | Improved |
| pretrain on Mikado, freeze first three encoder blocks | bicameral network · D2SA boundaries | AP: more than 10 points | Improved |
| pretrain on Mikado+ instead of Mikado with 12.5% of finetuning set | bicameral network · D2SA boundaries | AP: 3.4 -> 10.1 points | Improved |
Closest papers by meaning
- Robust Instance Segmentation Through Reasoning About Multi-Object Occlusion (CVPR 2021 · similarity 0.88)
- Occlusion-Aware Instance Segmentation Via BiLayer Network Architectures (TPAMI 2023 · similarity 0.88)
- Deep Occlusion-Aware Instance Segmentation With Overlapping BiLayers (CVPR 2021 · similarity 0.87)
- 3D-SIS: 3D Semantic Instance Segmentation of RGB-D Scans (CVPR 2019 · similarity 0.86)
- Learning to Generate Multiple Objects from Dense and Occluded Layouts (NeurIPS 2026 · similarity 0.85)
- Segment Anything, Even Occluded (CVPR 2025 · similarity 0.85)
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.
