Research map / Detection and segmentation / CNN features
SCNet: Training Inference Sample Consistency for Instance Segmentation
Thang Vu, Haeyong Kang, Chang D. Yoo
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 SCNet instead of HTC | ResNet-50 cascade instance segmentation · COCO test-dev | mask AP: 38.5 -> 40.2 | Improved |
| Use SCNet instead of HTC | ResNet-101 cascade instance segmentation · COCO test-dev | mask AP: 39.7 -> 41.3 | Improved |
| Use SCNet instead of HTC | ResNeXt-101 cascade instance segmentation · COCO test-dev | mask AP: 41.3 -> 42.7 | Improved |
| Add effective sample consistency (ablation) | Cascade Mask R-CNN with semantic branch · COCO val | mask AP: 37.4 -> 38.8 | Improved |
| Add effective sample consistency (ablation) | Cascade Mask R-CNN with semantic branch · COCO val | speed (fps): 4.5 -> 6.5 | Improved |
| Use effective sample consistency rather than naive sample consistency (ablation) | Cascade Mask R-CNN · COCO val | speed (fps): 4.5 -> 6.5 | Improved |
| Add feature relay (ablation) | Mask R-CNN · COCO val | mask AP: 35.1 -> 35.7 | Improved |
| Add feature relay (ablation) | Cascade Mask R-CNN · COCO val | mask AP: 37.4 -> 38.0 | Improved |
| Add feature relay (ablation) | Cascade Mask R-CNN · COCO val | speed (fps): 4.5 -> 4.1 | Hurt |
| Add global context (ablation) | Cascade Mask R-CNN with sample consistency and feature relay · COCO val | mask AP: 39.0 -> 39.8 | Improved |
| Increase global context loss weight from 0 to 1 (ablation) | SCNet · COCO val | mask AP: 39.6 -> 39.6 | No clear effect |
| Use global context loss weight 5 instead of 3 (ablation) | SCNet · COCO val | mask AP: 39.8 -> 39.6 | Hurt |
Later papers that compared with this method
5 later accepted papers report SC-Net as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Supervise Less, See More: Training-free Nuclear Instance Segmentation with Prototype-Guided Prompting (ICML 2026 · better on 2 of 2 shared results)
- Bridging RGB and Hematoxylin Components: An Interleaved Guidance and Fusion Framework for Point Supervised Nuclei Segmentation (CVPR 2026 · better on 1 of 2 shared results)
- DocSAM: Unified Document Image Segmentation via Query Decomposition and Heterogeneous Mixed Learning (CVPR 2025 · better on 0 of 1 shared result)
- Mudslide: A Universal Nuclear Instance Segmentation Method (CVPR 2024 · better on 4 of 4 shared results)
- Deformable Part Region Learning for Object Detection (AAAI 2022 · better on 1 of 2 shared results)
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
- Semantic Consistency Networks for 3D Object Detection (AAAI 2021 · similarity 0.87)
- S4Net: Single Stage Salient-Instance Segmentation (CVPR 2019 · similarity 0.85)
- Instance-Aware Semantic Segmentation via Multi-Task Network Cascades (CVPR 2016 · similarity 0.85)
- Rethinking ImageNet Pre-Training (ICCV 2019 · similarity 0.84)
- SCNN: A General Distribution Based Statistical Convolutional Neural Network with Application to Video Object Detection (AAAI 2019 · similarity 0.83)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · 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.
