Research map / Detection and segmentation / CNN features
DCT-Mask: Discrete Cosine Transform Mask Representation for 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 |
|---|---|---|---|
| Replace the binary mask grid with a compact DCT representation | Mask R-CNN with ResNet-50 · COCO val2017 | mask AP: 35.2 -> 36.5 | Helped |
| Replace the binary mask grid with a compact DCT representation | Mask R-CNN with ResNet-50 · LVIS* val | mask AP: 37.6 -> 39.7 | Helped |
| Replace the binary mask grid with a compact DCT representation | Mask R-CNN with ResNet-50 · Cityscapes val | AP: 33.0 -> 35.6 | Helped |
| Use DCT mask representation instead of the binary grid | Mask R-CNN with ResNeXt-101 · LVIS* val | mask AP: 42.1 -> 45.2 | Helped |
| Raise binary-grid resolution from 28×28 to 64×64 (ablation) | Mask R-CNN · COCO val2017 | mask AP: 35.2 -> 34.4 | Hurt |
| Raise binary-grid resolution from 28×28 to 128×128 (ablation) | Mask R-CNN · COCO val2017 | mask AP: 35.2 -> 32.9 | Hurt |
| Increase DCT mask resolution from 32×32 to 128×128 at 300 dimensions (ablation) | DCT-Mask R-CNN · COCO val2017 | mask AP: 35.4 -> 36.5 | Helped |
| Increase DCT mask resolution from 128×128 to 256×256 at 300 dimensions (ablation) | DCT-Mask R-CNN · COCO val2017 | mask AP: 36.5 -> 36.5 | No clear effect |
| Reduce the DCT vector dimension from 300 to 100 (ablation) | DCT-Mask R-CNN · COCO val2017 | mask AP: 36.5 -> 35.3 | Hurt |
| Increase the DCT vector dimension above 300 (ablation) | DCT-Mask R-CNN · COCO val2017 | mask AP: 36.5 -> 36.5 | No clear effect |
| Add three fully connected layers to the standard binary-grid mask branch (ablation) | Mask R-CNN · COCO val2017 | mask AP: 34.7 -> 34.6 | No clear effect |
| Apply L1 loss to the standard binary-grid mask branch (ablation) | Mask R-CNN with three FC layers · COCO val2017 | mask AP: 34.6 -> 18.5 | Hurt |
Closest papers by meaning
- Mask Encoding for Single Shot Instance Segmentation (CVPR 2020 · similarity 0.89)
- PatchDCT: Patch Refinement for High Quality Instance Segmentation (ICLR 2023 · similarity 0.88)
- BoxInst: High-Performance Instance Segmentation With Box Annotations (CVPR 2021 · similarity 0.87)
- Instance Segmentation With Mask-Supervised Polygonal Boundary Transformers (CVPR 2022 · similarity 0.87)
- Masked-Attention Mask Transformer for Universal Image Segmentation (CVPR 2022 · similarity 0.87)
- MaskDiff: Modeling Mask Distribution with Diffusion Probabilistic Model for Few-Shot Instance Segmentation (AAAI 2024 · similarity 0.87)
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.