Detect Any Mirrors: Boosting Learning Reliability on Large-Scale Unlabeled Data with an Iterative Data Engine
Accepted at CVPR 2025
This project introduces a robust method for detecting mirrors across diverse real-world scenes, leveraging large-scale unlabeled data with an iterative data engine.
You can download the dataset here:
👉 Download Dataset
- Content:
- 220,000 images from various scenarios involving mirrors.
- Corresponding pseudo labels.
The pre-trained model is available at the same link:
👉 Download Pre-trained Model
- After downloading, place the model weights inside the
weight/directory.
To perform inference with the pre-trained model, simply run:
python inference.py