Research map / Detection and segmentation / CNN features
Sparse Instance Activation for Real-Time 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 |
|---|---|---|---|
| Use sparse IAM-based instance features with single-level prediction and no NMS | SparseInst · COCO test-dev | mask AP | Helped |
| Add PPM to the encoder (ablation) | SparseInst · COCO val2017 | AP: 29.8 -> 31.3 | Helped |
| Add multi-scale feature fusion from P3 to P5 (ablation) | SparseInst with PPM · COCO val2017 | AP: 31.3 -> 32.0 | Helped |
| Add coordinate features (ablation) | SparseInst decoder · COCO val2017 | AP: 31.5 -> 32.0 | Helped |
| Replace the final branch convolutions with deformable convolutions (ablation) | SparseInst decoder · COCO val2017 | AP_L: 48.2 -> 49.2 | Helped |
| Increase decoder depth from 4 to 6 layers (ablation) | SparseInst decoder · COCO val2017 | AP: 32.0 -> 32.4 | Helped |
| Reduce decoder depth from 4 to 2 layers (ablation) | SparseInst decoder · COCO val2017 | AP: 32.0 -> 31.0 | Hurt |
| Reduce decoder width from 256 to 128 channels (ablation) | SparseInst decoder · COCO val2017 | AP: 32.0 -> 30.6 | Hurt |
| Replace sigmoid with softmax for IAM activations (ablation) | SparseInst IAM · COCO val2017 | AP: 32.0 -> 31.6 | Hurt |
| Replace the IAM 3×3 convolution with a 1×1 convolution (ablation) | SparseInst IAM · COCO val2017 | AP: 32.0 -> 30.8 | Hurt |
| Add an extra 3×3 convolution with ReLU to IAM (ablation) | SparseInst IAM · COCO val2017 | AP: 32.0 -> 31.9 | No clear effect |
| Use Group-IAM with 4 groups (ablation) | SparseInst IAM · COCO val2017 | AP: 32.0 -> 32.7 | Helped |
Closest papers by meaning
- Learning Instance Activation Maps for Weakly Supervised Instance Segmentation (CVPR 2019 · similarity 0.90)
- FastInst: A Simple Query-Based Model for Real-Time Instance Segmentation (CVPR 2023 · similarity 0.90)
- SOLOv2: Dynamic and Fast Instance Segmentation (NeurIPS 2020 · similarity 0.88)
- BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation (CVPR 2020 · similarity 0.88)
- Fully Convolutional Instance-Aware Semantic Segmentation (CVPR 2017 · similarity 0.87)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · 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.