Research map / Detection and segmentation / CNN features
Mask R-CNN
Kaiming He, Georgia Gkioxari, Piotr Dollar, Ross Girshick
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 per-pixel sigmoid with binary loss (decoupled masks) instead of softmax (ablation) | Mask R-CNN ResNet-50-C4 · COCO minival instance segmentation | mask AP: 24.8 -> 30.3 | Improved |
| predict class-specific masks instead of class-agnostic (ablation) | Mask R-CNN ResNet-50-C4 · COCO minival instance segmentation | mask AP: 29.7 -> 30.3 | No clear effect |
| replace RoIPool with RoIAlign (stride 16) (ablation) | Mask R-CNN ResNet-50-C4 · COCO minival instance segmentation | mask AP: 26.9 -> 30.3 | Improved |
| replace RoIWarp with RoIAlign (ablation) | Mask R-CNN ResNet-50-C4 · COCO minival instance segmentation | mask AP: 27.2 -> 30.2 | Improved |
| use average instead of max pooling in RoIAlign (ablation) | Mask R-CNN ResNet-50-C4 · COCO minival instance segmentation | mask AP: 30.2 -> 30.3 | No clear effect |
| replace RoIPool with RoIAlign (stride 32) (ablation) | Mask R-CNN ResNet-50-C5 · COCO minival instance segmentation (stride 32) | mask AP: 23.6 -> 30.9 | Improved |
| add RoIAlign with FPN (ablation) | Mask R-CNN FPN · COCO minival instance segmentation | mask AP: +1.5 mask AP | Improved |
| use an FCN mask branch instead of MLP (ablation) | Mask R-CNN ResNet-50-FPN · COCO minival instance segmentation | mask AP: 31.5 -> 33.6 | Improved |
| train with the mask branch (multi-task) instead of box-only (ablation) | Faster R-CNN, RoIAlign ResNet-101-FPN · COCO test-dev object detection | box AP: 37.3 -> 38.2 | Improved |
| replace ResNet-101-FPN with ResNeXt-101-FPN (ablation) | Mask R-CNN · COCO test-dev instance segmentation | mask AP: 35.7 -> 37.1 | Improved |
| replace ResNet-101-C4 with ResNet-101-FPN (ablation) | Mask R-CNN · COCO minival instance segmentation | mask AP: 32.7 -> 35.4 | Improved |
| replace RoIPool with RoIAlign for keypoints (ablation) | Mask R-CNN ResNet-50-FPN · COCO minival keypoint detection | keypoint AP: 59.8 -> 64.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Faster R-CNN — Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016) · better on 1 of 1 shared result
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
- Mask Scoring R-CNN (CVPR 2019 · similarity 0.91)
- Sparse R-CNN: End-to-End Object Detection With Learnable Proposals (CVPR 2021 · similarity 0.90)
- Mask Encoding for Single Shot Instance Segmentation (CVPR 2020 · similarity 0.90)
- Sparse R-CNN: An End-to-End Framework for Object Detection (TPAMI 2023 · similarity 0.90)
- Meta R-CNN: Towards General Solver for Instance-Level Low-Shot Learning (ICCV 2019 · similarity 0.90)
- BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation (CVPR 2020 · similarity 0.90)
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.
