Research map / Detection and segmentation / CNN features
Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up 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 |
|---|---|---|---|
| Use Adam instead of SGD momentum (ablation) | Panoptic-DeepLab · Cityscapes val | PQ: +0.7% PQ | Helped |
| Use MSE instead of sigmoid cross entropy for the center heatmap (ablation) | Panoptic-DeepLab · Cityscapes val | PQ: +0.8% PQ | Helped |
| Use dual decoder and dual ASPP (ablation) | Panoptic-DeepLab · Cityscapes val | PQ: 61.8 -> 62.5 | Helped |
| Train with a large crop of 1025×2049 instead of 513×1025 (ablation) | Panoptic-DeepLab · Cityscapes val | AP and mIoU: +0.6% AP; +0.9% mIoU | Helped |
| Increase semantic branch channels from 128 to 256 (ablation) | Panoptic-DeepLab · Cityscapes val | PQ: 62.7 -> 63.0 | Helped |
| Increase instance branch channels from 128 to 256 (ablation) | Panoptic-DeepLab · Cityscapes val | PQ: 63.0 -> 62.1 | Hurt |
| Train only the semantic segmentation task (ablation) | Panoptic-DeepLab · Cityscapes val | mIoU: 80.5 -> 80.3 | No clear effect |
| Add left-right flipped inference (ablation) | Panoptic-DeepLab · Cityscapes val | PQ: 63.0 -> 63.4 | Helped |
| Add multiscale inference to flipped inference (ablation) | Panoptic-DeepLab · Cityscapes val | PQ: 63.4 -> 64.1 | Helped |
| Add atrous convolution in the last two backbone blocks at inference (ablation) | Panoptic-DeepLab · Cityscapes val | PQ: no improvement over PQ | No clear effect |
| Remove the stride that creates the 1/32 feature map (ablation) | Auto-DeepLab · Mapillary Vistas val | PQ: 1% PQ | Helped |
| Add atrous convolution at rate 2 in HRNet+'s output-stride-32 branch (ablation) | HRNet-W48+ · Mapillary Vistas val | PQ: 40.6 -> 40.5 | Hurt |
Closest papers by meaning
- Unifying Training and Inference for Panoptic Segmentation (CVPR 2020 · similarity 0.91)
- Combinatorial Optimization for Panoptic Segmentation: A Fully Differentiable Approach (NeurIPS 2021 · similarity 0.90)
- UPSNet: A Unified Panoptic Segmentation Network (CVPR 2019 · similarity 0.90)
- LPSNet: A Lightweight Solution for Fast Panoptic Segmentation (CVPR 2021 · similarity 0.90)
- Scene-Centric Unsupervised Panoptic Segmentation (CVPR 2025 · similarity 0.88)
- Fully Data-Driven Pseudo Label Estimation for Pointly-Supervised Panoptic Segmentation (AAAI 2024 · 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.