Research map / Detection and segmentation / CNN features
TensorMask: A Foundation for Dense Object Segmentation
Xinlei Chen, Ross Girshick, Kaiming He, Piotr Dollar
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 aligned instead of natural representation with upscaling lambda=5 (ablation) | TensorMask upscaling head · COCO val | mask AP: 19.2 -> 28.4 | Improved |
| use aligned instead of natural representation with lambda=3 (ablation) | TensorMask upscaling head · COCO val | mask AP: 24.7 -> 28.8 | Improved |
| use aligned instead of natural representation with lambda=1.5 (ablation) | TensorMask upscaling head · COCO val | mask AP: 28.0 -> 28.9 | Improved |
| use aligned instead of natural representation without upscaling (ablation) | TensorMask simple head · COCO val | mask AP: 28.5 -> 28.9 | No clear effect |
| use bilinear instead of nearest-neighbor interpolation, lambda=5 (ablation) | aligned upscaling head · COCO val | mask AP: 25.3 -> 28.4 | Improved |
| use bilinear instead of nearest-neighbor interpolation, lambda=1.5 (ablation) | aligned upscaling head · COCO val | mask AP: 28.6 -> 28.9 | Improved |
| use tensor bipyramid instead of feature pyramid (ablation) | TensorMask · COCO val | mask AP: 28.9 -> 34.0 | Improved |
| use tensor bipyramid instead of feature pyramid (ablation) | TensorMask · COCO val | mask AP_L: 40.7 -> 48.4 | Improved |
| add a second window size 11x11 (ablation) | TensorMask tensor bipyramid · COCO val | mask AP: 34.0 -> 35.2 | Improved |
| add scale augmentation and longer 72-epoch schedule | Mask R-CNN · COCO test-dev instance segmentation | mask AP: 34.9 -> 36.8 | Improved |
| use TensorMask instead of Mask R-CNN (R-50) | dense sliding-window segmentation · COCO test-dev instance segmentation | mask AP: 36.8 -> 35.4 | Hurt |
| use TensorMask instead of Mask R-CNN (R-101) | dense sliding-window segmentation · COCO test-dev instance segmentation | mask AP: 38.3 -> 37.1 | Hurt |
Later papers that compared with this method
14 later accepted papers report TensorMask as a baseline in their results tables; 12 of them report a better number on most of the results they share with it.
- PointINS: Point-Based Instance Segmentation (TPAMI 2021 · better on 2 of 2 shared results)
- SOTR: Segmenting Objects With Transformers (ICCV 2021 · better on 1 of 1 shared result)
- SOLO: A Simple Framework for Instance Segmentation (TPAMI 2021 · better on 1 of 1 shared result)
- PolarMask++: Enhanced Polar Representation for Single-Shot Instance Segmentation and Beyond (TPAMI 2021 · better on 1 of 1 shared result)
- BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation (CVPR 2020 · better on 2 of 2 shared results)
- SipMask: Spatial Information Preservation for Fast Image and Video Instance Segmentation (ECCV 2020 · better on 2 of 2 shared results)
- SOLO: Segmenting Objects by Locations (ECCV 2020 · better on 2 of 2 shared results)
- CenterMask: Real-Time Anchor-Free Instance Segmentation (CVPR 2020 · better on 1 of 1 shared result)
6 more not listed.
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
- SOLO: A Simple Framework for Instance Segmentation (TPAMI 2021 · similarity 0.89)
- SipMask: Spatial Information Preservation for Fast Image and Video Instance Segmentation (ECCV 2020 · similarity 0.89)
- BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation (CVPR 2020 · similarity 0.89)
- ShapeMask: Learning to Segment Novel Objects by Refining Shape Priors (ICCV 2019 · similarity 0.89)
- Mask R-CNN (ICCV 2017 · similarity 0.89)
- Semi-convolutional Operators for Instance Segmentation (ECCV 2018 · similarity 0.88)
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.
