Research map / Detection and segmentation / CNN features
SG-Net: Spatial Granularity Network for One-Stage Video Instance Segmentation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace the two-stage MaskTrack R-CNN pipeline with one-stage SG-Net | VIS with ResNet-50 · YouTube-VIS validation | AP: 30.3 -> 34.8 | Helped |
| Use SG-Net instead of SipMask | VIS with ResNet-101 · YouTube-VIS validation | AP: 35.8 -> 36.3 | Helped |
| Replace MaskTrack R-CNN with SG-Net | VIS with ResNet-101 · YouTube-VIS validation | Time (ms): 142.8 -> 50.6 | Helped |
| Add the base mask module with multi-level P3-P5 features (ablation) | SG-Net baseline · YouTube-VIS validation | AP: 31.1 -> 34.3 | Helped |
| Add the spatial attention module (ablation) | SG-Net baseline · YouTube-VIS validation | AP: 31.1 -> 35.2 | Helped |
| Add both the base mask and spatial attention modules (ablation) | SG-Net baseline · YouTube-VIS validation | AP: 31.1 -> 36.1 | Helped |
| Add the GIoU loss to the model with base mask and spatial attention modules (ablation) | SG-Net · YouTube-VIS validation | AP: 36.1 -> 36.3 | Helped |
| Increase maximum box divisions from 2×* to 6×* (ablation) | SG-Net spatial attention module · YouTube-VIS validation | AP: 34.0 -> 36.3 | Helped |
| Increase maximum box divisions from 6×* to 8×* (ablation) | SG-Net spatial attention module · YouTube-VIS validation | AP: 36.3 -> 36.4 | No clear effect |
| Increase maximum box divisions from 6×* to 8×* (ablation) | SG-Net spatial attention module · YouTube-VIS validation | Time (ms): 50.6 -> 88.6 | Hurt |
| Upsample base mask features by factor 4 instead of using no upsampling (ablation) | SG-Net · YouTube-VIS validation | AP: 34.9 -> 36.3 | Helped |
| Increase base mask feature upsampling from factor 2 to factor 4 (ablation) | SG-Net · YouTube-VIS validation | AP@0.75: 38.6 -> 39.6 | Helped |
Closest papers by meaning
- End-to-End Video Instance Segmentation With Transformers (CVPR 2021 · similarity 0.89)
- Mask-Free Video Instance Segmentation (CVPR 2023 · similarity 0.88)
- VISOLO: Grid-Based Space-Time Aggregation for Efficient Online Video Instance Segmentation (CVPR 2022 · similarity 0.88)
- Video K-Net: A Simple, Strong, and Unified Baseline for Video Segmentation (CVPR 2022 · similarity 0.88)
- Multi-Granularity Video Object Segmentation (AAAI 2025 · similarity 0.86)
- Robust and Consistent Online Video Instance Segmentation via Instance Mask Propagation (AAAI 2025 · similarity 0.86)
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.