Research map / Detection and segmentation / CNN features
CMT-DeepLab: Clustering Mask Transformers for Panoptic 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 standard query updating with clustering-style center updates (ablation) | CMT-DeepLab · COCO ablation | PQ: 46.2 -> 47.1 | Helped |
| Use pixel-wise instead of instance-wise contrastive loss (ablation) | CMT-DeepLab · COCO ablation | PQ: 47.1 -> 47.5 | Helped |
| Add reference mask prediction (ablation) | CMT-DeepLab · COCO ablation | PQ: 46.2 -> 46.6 | Helped |
| Add coordinate convolution to reference-mask features (ablation) | CMT-DeepLab · COCO ablation | PQ: 46.6 -> 46.9 | Helped |
| Use the stronger six-layer decoder (ablation) | CMT-DeepLab · COCO ablation | PQ: 46.2 -> 47.1 | Helped |
| Add clustering updates to the stronger decoder (ablation) | CMT-DeepLab · COCO ablation | PQ: 47.1 -> 48.1 | Helped |
| Use ImageNet-22K rather than ImageNet-1K pretraining (ablation) | CMT-DeepLab · COCO ablation | PQ: 48.4 -> 49.3 | Helped |
| Use RFN (ablation) | CMT-DeepLab · COCO ablation | PQ: 49.3 -> 50.1 | Helped |
| Use mask-wise merge instead of per-pixel post-processing (ablation) | CMT-DeepLab · COCO ablation | PQ: 50.1 -> 50.6 | Helped |
| Increase training from 100k to 200k iterations (ablation) | CMT-DeepLab · COCO ablation | PQ: 50.1 -> 50.6 | Helped |
| Increase input resolution from 641 to 1281 (ablation) | CMT-DeepLab · COCO ablation | PQ: 50.1 -> 53.0 | Helped |
| Use Axial-R104 instead of Axial-R50 (ablation) | CMT-DeepLab · COCO ablation | PQ: 52.2 -> 54.1 | Helped |
Closest papers by meaning
- MaX-DeepLab: End-to-End Panoptic Segmentation With Mask Transformers (CVPR 2021 · similarity 0.89)
- Masked-Attention Mask Transformer for Universal Image Segmentation (CVPR 2022 · similarity 0.87)
- ClusterFomer: Clustering As A Universal Visual Learner (NeurIPS 2023 · similarity 0.87)
- Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation (CVPR 2020 · similarity 0.87)
- Panoptic SegFormer: Delving Deeper Into Panoptic Segmentation With Transformers (CVPR 2022 · similarity 0.86)
- PEM: Prototype-based Efficient MaskFormer for Image Segmentation (CVPR 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.