yinkejia/iPhone-PTZ

This dataset contains processed benchmarks for both iPhone and iPhone-PTZ.

CamDirector: Towards Long-Term Coherent Video Trajectory Editing
Zhihao Shi^1*, Kejia Yin^2*, Weilin Wan³, Yuhongze Zhou⁴, Yuanhao Yu¹, Xinxin Zuo⁵, Qiang Sun^2,6, Juwei Lu²
McMaster, UofT, HKU, McGill, Concordia, MBZUAI
^* Equal contribution
CVPR 2026

Dataset Structure

After unzip the zip files you will get two folders below:

./evaluation_benchmarks
├── iPhone/
│   ├── apple
|       ├── camera_poses
│           ├── source_c2ws.pt
│           ├── source_Ks.pt
│           ├── target_c2ws.pt
│           └── target_Ks.pt
|       ├── depth_maps
│           ├── frame_00000_depth.png
│           ├── frame_00001_depth.png
│           ├── frame_00002_depth.png
│           └── ...
|       ├── point_clouds
│           ├── sparse_pcds_src.saftetensors
|       ├── source_imgs
│           ├── frame_00000.png
│           ├── frame_00001.png
│           ├── frame_00002.png
│           └── ...
|       ├── target_imgs
│           ├── frame_00000.png
│           ├── frame_00001.png
│           ├── frame_00002.png
│           └── ...
|   ├── block
|   ├── ...
├── iPhone-PTZ/
│   ├── container
|       ├── camera_poses
│           ├── source_c2ws.pt
│           ├── source_Ks.pt
│           ├── target_c2ws.pt
│           └── target_Ks.pt
|       ├── depth_maps
│           ├── frame_00000_depth.png
│           ├── frame_00001_depth.png
│           ├── frame_00002_depth.png
│           └── ...
|       ├── point_clouds
│           ├── sparse_pcds_src.saftetensors
|       ├── source_imgs
│           ├── frame_00000.png
│           ├── frame_00001.png
│           ├── frame_00002.png
│           └── ...
|       ├── target_imgs
│           ├── frame_00000.png
│           ├── frame_00001.png
│           ├── frame_00002.png
│           └── ...
|   ├── corner
|   ├── ...

We provide our processed data for both iPhone and iPhone-PTZ benchmarks. For each scene, we provide the following:

• Camera poses (c2ws and Ks) for source and target video frames in opencv camera convention.
• Depth maps for source and target video frames, please refer to vis_benchmark.py on how to read them
• Sparse points clouds for source video frames. If you prefer use your own depth, you may use these sparse point clouds to align your depths to our camera system.

Results Structure

./output_frames
├── camdiretor/
│   ├── apple
│       ├── frame_00000.png
│       ├── frame_00002.png
│       ├── frame_00004.png
│       └── ...
|   ├── block
|   ├── ...
|   ├── container
|   ├── corner
|   ├── ...

Please refer to iPhone-PTZ-benchmark for details.