Add dataset card for GemDepth evaluation datasets

README.md

@@ -1,3 +1,64 @@
----
-license: apache-2.0
----
+---
+license: apache-2.0
+task_categories:
+- depth-estimation
+---
+
+# GemDepth Evaluation Datasets
+
+This repository contains the evaluation datasets for **GemDepth**, as presented in the paper [GemDepth: Geometry-Embedded Features for 3D-Consistent Video Depth](https://huggingface.co/papers/2605.10525).
+
+[**Project Page**](https://r11031.github.io/website/) | [**GitHub**](https://github.com/Yuecheng919/GemDepth) | [**Paper**](https://huggingface.co/papers/2605.10525)
+
+## Introduction
+GemDepth is a framework for 3D-consistent video depth estimation that leverages explicit awareness of camera motion and global 3D structure. This repository hosts the processed benchmarks used for zero-shot evaluation, including Sintel, KITTI, Bonn, and ScanNet.
+
+## Dataset Structure
+The datasets are provided as compressed archives:
+- `sintel.tar.gz`
+- `kitti.tar.gz`
+- `bonn.tar.gz`
+- `scannet.tar.gz`
+
+## Usage
+
+### Preparation
+To extract the datasets after downloading, you can use the extraction scripts provided in the official GitHub repository:
+
+```bash
+pip install natsort
+cd dataset/dataset_extract
+# Replace ${dataset} with sintel, kitti, bonn, or scannet
+python dataset_extrtact${dataset}.py
+```
+*Note: This script extracts the data and generates the JSON files required for evaluation.*
+
+### Run Inference
+To evaluate a model on these datasets:
+
+```bash
+python evaluation/inference/infer/infer.py \
+    --infer_path ${out_path} \
+    --json_file ${json_path} \
+    --datasets ${dataset}
+```
+
+### Run Evaluation
+To calculate metrics:
+
+```bash
+python evaluation/eval/eval.py \
+    --infer_path ${pred_root} \
+    --benchmark_path ${benchmark_root} \
+    --datasets ${dataset}
+```
+
+## Citation
+```bibtex
+@article{liu2026gemdepth,
+  title={GemDepth: Geometry-Embedded Features for 3D-Consistent Video Depth},
+  author={Yuecheng liu and Junda Cheng and Longliang Liu and Wenjing Liao and Hanrui Cheng and Yuzhou Wang and Xin Yang},
+  journal={arXiv preprint arXiv:2605.10525},
+  year={2026}
+}
+```