Research map / Detection and segmentation / CNN features
YOLACT++ Better Real-Time Instance Segmentation
Daniel Bolya, Chong Zhou, Fanyi Xiao, Yong Jae Lee
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 |
|---|---|---|---|
| replace fc-layer mask outputs with prototype and coefficient assembly (ablation) | one-stage detector · COCO val2017 | mask AP: 20.7 -> 29.9 | Improved |
| use Fast NMS instead of standard NMS (ablation) | YOLACT · COCO val2017 | mask AP and FPS: 30.0 -> 29.9 AP, 24.0 -> 33.5 FPS | Mixed |
| increase prototypes k from 8 to 32 (ablation) | YOLACT · COCO val2017 | mask AP: 26.8 -> 27.7 | Improved |
| increase prototypes k from 32 to 256 (ablation) | YOLACT · COCO val2017 | mask AP: 27.7 -> 27.7 | No clear effect |
| add semantic segmentation loss on P3 during training (ablation) | YOLACT · COCO | mask mAP: +0.4 mAP | Improved |
| add 3 scales per FPN level to anchors (ablation) | YOLACT · COCO val2017 | mask AP: 27.7 -> 30.2 | Improved |
| add 5 aspect ratios to anchors (ablation) | YOLACT · COCO val2017 | mask AP: 27.7 -> 28.0 | Improved |
| add deformable convolution with interval 3 (ablation) | YOLACT++ R-101 · COCO val2017 | mask AP: 31.7 -> 33.3 | Improved |
| use interval 3 instead of DCN in all blocks (ablation) | YOLACT++ R-101 · COCO val2017 | mask AP: 33.5 -> 33.3 | No clear effect |
| add fast mask re-scoring (ablation) | YOLACT++ R-101 · COCO val2017 | mask AP: 33.3 -> 34.4 | Improved |
| add fast mask re-scoring (ablation) | YOLACT++ R-50 · COCO val2017 | mask AP: 32.7 -> 33.7 | Improved |
| replace Fast NMS with standard NMS in Mask R-CNN (ablation) | Mask R-CNN · COCO val2017 | mask AP and FPS: 36.1 -> 35.8 AP, 8.6 -> 9.9 FPS | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- PA-Net — PANet: Few-Shot Image Semantic Segmentation With Prototype Alignment (ICCV 2019) · 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
- YOLACT: Real-Time Instance Segmentation (ICCV 2019 · similarity 0.97)
- FastInst: A Simple Query-Based Model for Real-Time Instance Segmentation (CVPR 2023 · similarity 0.90)
- BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation (CVPR 2020 · similarity 0.89)
- Real-Time Instance Segmentation With Discriminative Orientation Maps (ICCV 2021 · similarity 0.89)
- Sparse Instance Activation for Real-Time Instance Segmentation (CVPR 2022 · similarity 0.89)
- SOLO: A Simple Framework for Instance Segmentation (TPAMI 2021 · similarity 0.89)
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.
