rankseg 0.0.3

RankSEG: A Statistically Consistent Segmentation Prediction Solver for Dice and IoU Metrics Optimization

🧩 RankSEG

🚀 RankSEG: Boost Segmentation Metrics without Retraining

Improve your existing segmentation models performance in Dice/IoU metrics instantly.

RankSEG is a plug-and-play post-processing module that improves segmentation results during inference. It works with ANY pre-trained logit/prob-outcome segmentation model (SAM, DeepLab, SegFormer, UPerNet, etc.) without any retraining or fine-tuning.

Instead of using simple thresholding or argmax (which don't care about Dice/IoU scores), RankSEG directly optimizes for these metrics - giving you better results without any extra training.

Explore RankSEG by reading our documentation.

---

⚡ Quick Start

Get started in seconds. RankSEG is designed to be dropped into your existing inference pipeline.

1. Installation

pip install -U rankseg

2. Optimize Predictions

Add 3 lines of code to your inference loop:

import torch
import torch.nn.functional as F
from rankseg import RankSEG

# 1. Initialize RankSEG (optimizing for Dice score)
rankseg = RankSEG(metric='dice', solver='RMA')

# 2. Get your model's probability outputs (Batch, Class, Height, Width)
# Example: probs = model(images).softmax(dim=1)
probs = F.softmax(torch.randn(4, 21, 256, 256), dim=1)

# 3. Get optimized predictions (No retraining needed!)
preds = rankseg.predict(probs)

💡 Try it now:

---

✨ Key Features

• 🚀 Instant Metric Boost: Consistently improves mIoU and mDice scores over standard argmax.
• 🔌 Plug-and-Play: Compatible with any PyTorch segmentation model.
• 🆓 Training-Free: Purely post-processing. No gradients, no backprop, no dataset needed.
• ⚡ Efficient: Optimized solvers (RMA) for real-time inference.
• 🧩 Flexible: Supports multi-class and multi-label segmentation.

---

📊 Why RankSEG?

Standard segmentation methods use argmax or thresholding, which are not optimized for evaluation metrics like Dice or IoU. RankSEG solves this by directly optimizing the target metric during inference.

Performance Comparison (No Retraining):

Model	Dataset	mIoU (Argmax)	mIoU (RankSEG)	mDice (Argmax)	mDice (RankSEG)
DeepLabV3+ (ResNet101)	PASCAL VOC	77.25%	78.14% ↑0.89%	82.08%	83.14% ↑1.06%
SegFormer (MiT-B4)	PASCAL VOC	77.57%	78.59% ↑1.02%	82.15%	83.22% ↑1.07%
UPerNet (ConvNeXt)	PASCAL VOC	79.52%	80.31% ↑0.79%	84.11%	84.98% ↑0.87%
PSPNet (ResNet101)	Cityscapes	65.89%	66.53% ↑0.64%	73.55%	74.28% ↑0.73%
DeepLabV3+ (ResNet101)	Cityscapes	66.17%	66.68% ↑0.51%	73.71%	74.33% ↑0.62%
UPerNet (ConvNeXt)	Cityscapes	68.83%	69.57% ↑0.74%	76.08%	76.97% ↑0.89%
SegFormer (MiT-B4)	ADE20K	40.00%	40.82% ↑0.82%	46.50%	47.57% ↑1.07%
UPerNet (ConvNeXt)	ADE20K	42.86%	43.84% ↑0.98%	49.61%	50.85% ↑1.24%
CPT (Swin-Large)	ADE20K	44.59%	45.56% ↑0.97%	51.27%	52.58% ↑1.31%

Results from our NeurIPS 2025 paper.

---

🛠️ Integrations

RankSEG works out-of-the-box with any PyTorch-based segmentation framework.

Framework	Task	Integration Guide
PyTorch (Native)	Semantic Seg.
SegmentAnything	Semantic Seg.
MMSegmentation	Semantic Seg.	Coming Soon
PaddleSeg	Semantic Seg.	Coming Soon

Note: Don't see your favorite framework? Open an issue or submit a PR!

---

📚 Documentation & Resources

• Full Documentation: rankseg.readthedocs.io
• JMLR Paper: RankSEG: A Consistent Ranking-based Framework for Segmentation
• NeurIPS Paper: RankSEG-RMA: An Efficient Segmentation Algorithm via Reciprocal Moment Approximation

🔗 Citation

If you use RankSEG in your research, please cite our papers:

@article{dai2023rankseg,
  title={RankSEG: A Consistent Ranking-based Framework for Segmentation},
  author={Dai, Ben and Li, Chunlin},
  journal={Journal of Machine Learning Research},
  volume={24},
  number={224},
  pages={1--50},
  year={2023}
}

@inproceedings{wang2025rankseg,
  title={RankSEG-RMA: An Efficient Segmentation Algorithm via Reciprocal Moment Approximation},
  author={Wang, Zixun and Dai, Ben},
  booktitle={Advances in Neural Information Processing Systems},
  year={2025}
}

🤝 Contributing

We welcome contributions! Please feel free to submit issues or pull requests on our GitHub repository.

If you find RankSEG useful, please give it a star! ⭐