Update README.md

README.md

@@ -1 +1,46 @@
-# UniTSFace
+## Introduction
+
+**This is the official PyTorch implementation of the ICCV 2023 paper.**
+
+[UniTSFace: Unified Threshold Integrated Sample-to-Sample Loss for Face Recognition.pdf]()
+
+[Supplementary.pdf]()
+
+
+## Get started
+
+1. **Prepare dataset**
+
+    Download [CASIA-Webface](https://drive.google.com/file/d/1KxNCrXzln0lal3N4JiYl9cFOIhT78y1l/view?usp=sharing) preprocessed by [insightface](https://github.com/deepinsight/insightface/blob/master/recognition/_datasets_/README.md).
+    ```console
+    unzip faces_webface_112x112.zip
+    ```
+
+2. **Train model**
+
+    Modify the 'data_path' in [train.py](train.py) (Line 56)
+
+    Select and uncomment the 'sample_to_sample_loss' in [backbone.py](backbone.py) (Line 71)
+    ```console
+    python train.py
+    ```
+
+4. **Test model**
+    ```console
+    python pytorch2onnx.py
+    zip model.zip model.onnx
+    ```
+    Upload model.zip to [MFR Ongoing](http://iccv21-mfr.com/#/leaderboard/academic) and then wait for the results.
+
+    We provide a pre-trained model (ResNet-50) on [Google Drive](https://drive.google.com/file/d/167zN2NYowc6UyP4CjwPfgW3xM86oUrWD/view?usp=drive_link) for easy and direct development. This model is trained on CASIA-WebFace and achieved 50.25% on MR-All and 99.53% on LFW.
+
+## Citation
+
+If you find **UniTSFace** useful in your research, please consider to cite:
+
+  ```bibtex
+  @InProceedings{NeurIPS_2023,
+    author    = {Zhou, Jiancan and Jia, Xi and Li, Qiufu and Shen, Linlin and Duan, Jinming},
+    title     = {UniTSFace: Unified Threshold Integrated Sample-to-Sample Loss for Face Recognition}
+  }
+  ```