Spider: A Unified Framework for Context-dependent Concept Segmentation

ICML, 2024
Xiaoqi Zhao* · Youwei Pang* . Wei Ji* . Baicheng Sheng . Jiaming Zuo · Lihe Zhang* · Huchuan Lu

Context-independent (CI) Concept vs. Context-dependent (CD) Concept

CD Concept Segmentation Survey Paper

(IJCV 2024) Towards Diverse Binary Segmentation via A Simple yet General Gated Network

Unified 8 CD Concept Segmentation Tasks

Spider: UniCDSeg Framework (You only train and infer once! 100% Unified Parameters.)

Performance

Potential

Continual/Zero-shot/Incremental Zero-shot learning

In-Context Learning

Datasets

• DUTS (SOD): Google Drive
• COD10K (COD): Google Drive
• SBU (SD): Google Drive
• Trans10K (TOS): Trans10K Website
• Five datasets (CPS): Google Drive
• COVID-19 data (COD): Google Drive
• BUSI (BLS): Google Drive
• ISIC18 (SLS): Google Drive

Trained Models

• Spider-ConvNext-B Google Drive
• Spider-ConvNext-L GitHub Release
• Spider-Swin-B Google Drive
• Spider-Swin-L GitHub Release

Prediction Maps

• Spider-Swin-B Google Drive
• Spider-Swin-L Google Drive
• Spider-ConvNeXt-B Google Drive
• Spider-ConvNeXt-L Google Drive

Evaluation Tools

• https://github.com/Xiaoqi-Zhao-DLUT/PySegMetric_EvalToolkit

To Do List

• Release data sets.
• Release model code.
• Release model weights.
• Release model prediction maps.

Citation

If you think Spider-UniCDSeg codebase are useful for your research, please consider referring us:

@inproceedings{Spider,
  title={Spider: A Unified Framework for Context-dependent Concept Segmentation},
  author={Zhao, Xiaoqi and Pang, Youwei and Ji, Wei and Sheng, Baicheng and Zuo, Jiaming and Zhang, Lihe and Lu, Huchuan},
  booktitle={ICML},
  year={2024}