diff --git a/README.md b/README.md
index e5da005..c1030d7 100644
--- a/README.md
+++ b/README.md
@@ -3,7 +3,12 @@
This is the official repository for STHN: Deep Homography Estimation for UAV Thermal Geo-localization with Satellite Imagery.
```
-bibtex TBD
+@article{xiao2024sthn,
+ title={STHN: Deep Homography Estimation for UAV Thermal Geo-localization with Satellite Imagery},
+ author={Xiao, Jiuhong and Zhang, Ning and Tortei, Daniel and Loianno, Giuseppe},
+ journal={arXiv preprint arXiv:2405.20470},
+ year={2024}
+}
```
**Developer: Jiuhong Xiao
Affiliation: [NYU ARPL](https://wp.nyu.edu/arpl/)
@@ -112,6 +117,26 @@ For training and evaluating the image-matching baselines (anyloc and STGL), plea
## Pretrained Models
Download pretrained models for $W_S=1536$ and $D_C=512$ m: [Download](https://drive.google.com/drive/folders/19Z0wqeDlJyzTZB1qc355G2WEww7I9rTB?usp=sharing)
+## Additional Details
+
+ Train/Val/Test split
+ Below is the visualization of the train-validation-test regions. The dataset includes thermal maps from six flights: three flights (conducted at 9 PM, 12 AM, and 2 AM) cover the upper region, and the other three flights (conducted at 10 PM, 1 AM, and 3 AM) cover the lower region. The lower region is further divided into training and validation subsets. The synthesized thermal images span a larger area (23,744m x 9,088m) but exclude the test region to assess generalization performance properly.
+
+ 
+
+
+
+ Architecture Details
+
+
+
+
+
+ Direct Linear Transformation Details
+ In practice, we use kornia's implementation [kornia](https://kornia.readthedocs.io/en/stable/geometry.transform.html#kornia.geometry.transform.get_perspective_transform).
+ For more details, you can refer to [wiki](https://en.wikipedia.org/wiki/Direct_linear_transformation).
+
+
## Acknowledgement
Our implementation refers to the following repositories and appreciate their excellent work.