Research map / Detection and segmentation / CNN features
SOLQ: Segmenting Objects by Learning Queries
Read the paperCheck 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 unified query representation (UQR) instead of separate query representation (SQR) with FPN mask branch (ablation) | D-DETR* R50 · COCO val | box AP: 45.6 -> 47.8 | Helped |
| use UQR instead of SQR (ablation) | D-DETR* R50 · COCO val | mask AP: 32.2 -> 39.5 | Helped |
| use UQR instead of SQR (ablation) | D-DETR* R101 · COCO val | box AP: 46.4 -> 48.3 | Helped |
| use DCT instead of flattening mask (ablation) | SOLQ · COCO val | mask AP: 29.7 -> 32.6 | Helped |
| use DCT instead of PCA (ablation) | SOLQ · COCO val | mask AP: 31.6 -> 32.6 | Helped |
| use DCT instead of sparse coding (ablation) | SOLQ · COCO val | mask AP: 11.3 -> 32.6 | Helped |
| raise mask vector loss weight from 0.3 to 3 (ablation) | SOLQ · COCO val | mask AP: 25.2 -> 29.7 | Helped |
| raise mask vector loss weight from 3 to 4 (ablation) | SOLQ · COCO val | mask AP: 29.7 -> 29.6 | No clear effect |
| add auxiliary mask vector loss on all 6 decoder layers instead of last only (ablation) | SOLQ · COCO val | mask AP: 29.7 -> 32.6 | Helped |
| raise mask resolution N from 64 to 128 (ablation) | SOLQ · COCO val | mask AP: 30.8 -> 32.6 | Helped |
| raise mask resolution N from 128 to 256 (ablation) | SOLQ · COCO val | mask AP: 32.6 -> 32.3 | No clear effect |
| raise mask vector dimension from 256 to 400 (ablation) | SOLQ · COCO val | mask AP: 32.6 -> 30.9 | Hurt |
Closest papers by meaning
- SOIT: Segmenting Objects with Instance-Aware Transformers (AAAI 2022 · similarity 0.88)
- Learning Dynamic Query Combinations for Transformer-based Object Detection and Segmentation (ICML 2023 · similarity 0.87)
- Boosting Low-Data Instance Segmentation by Unsupervised Pre-Training With Saliency Prompt (CVPR 2023 · similarity 0.86)
- LQMFormer: Language-aware Query Mask Transformer for Referring Image Segmentation (CVPR 2024 · similarity 0.86)
- Learning Equivariant Segmentation with Instance-Unique Querying (NeurIPS 2022 · similarity 0.85)
- MaskDiff: Modeling Mask Distribution with Diffusion Probabilistic Model for Few-Shot Instance Segmentation (AAAI 2024 · similarity 0.84)
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.