Title: Learning Dual-Stream Conditional Concepts in Compositional Zero-Shot Learning
Authors: Qingsheng Wang, Lingqiao Liu, Chenchen Jing, et.al.
Publication: IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE TPAMI), Vol. 47, No. 11, November 2025.
Published Paper: [https://ieeexplore.ieee.org/abstract/document/11141706/]
Thank you for reading our work.
We provide the pre-trained files for UT-Zappos50K, MIT-States, C-GQA, and VAW-CZSL in Google Drive. All pre-trained files are based on the frozen pre-trained OpenAI CLIP model in default FP16 format. Please run evaluation.py to evaluate with the following downloaded .pt files. You may get slightly different results from the reported ones due to your hardwares and environments.
| Dataset | Top-k | Seen | Unseen | HM | AUC | Size | URL |
|---|---|---|---|---|---|---|---|
| UT-Zappos50K | Top-1 | 70.1 | 76.2 | 58.8 | 46.2 | 318.7MB | Download |
| MIT-States | Top-1 | 49.2 | 52.4 | 38.7 | 21.8 | 319.7MB | Download |
| C-GQA | Top-1 | 42.6 | 34.3 | 30.1 | 12.7 | 321.7MB | Download |
| VAW-CZSL | Top-5 | 42.5 | 50.0 | 32.6 | 15.6 | 321.4MB | Download |
The downloaded .pt files are supposed to be placed in saves/(the chosen dataset)/XXX.pt. For example, ut-zap50k.pt should be placed like saves/ut-zap50k/ut-zap50k.pt.
If you find this work interesting please cite:
@ARTICLE{11141706,
author={Wang, Qingsheng and Liu, Lingqiao and Jing, Chenchen and Wang, Peng and Zhang, Yanning and Shen, Chunhua},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
title={Learning Dual-Stream Conditional Concepts in Compositional Zero-Shot Learning},
year={2025},
volume={47},
number={11},
pages={10076-10093},
keywords={Visualization;Semantics;Image recognition;Zero shot learning;Streaming media;Dairy products;Feature extraction;Computational modeling;Benchmark testing;Training;Compositional zero-shot learning;compositional generalization;tuning soft prompts;zero-shot learning},
doi={10.1109/TPAMI.2025.3597668}}