Download

---

Our dataset is available on both Hugging Face and Google Drive. We recommend using Hugging Face due to rate limits on Google Drive.

Hugging Face

Please follow the instructions on the Hugging Face repository to download the dataset.

Google Drive

The dataset has a validation and test split:

• val.zip – Contains all validation data
• test.zip – Contains all test sequences

Instructions on how to unzip the files are provided below.

Dataset Categories

The dataset includes three main types of scenes, grouped under the top-level Benchmark/ folder:

• Object Scanning (285 videos): Videos focus on individual objects across a wide variety of indoor and outdoor environments, lighting conditions, and camera trajectories. These include static and dynamic scenes, non-Lambertian surfaces, and texture-less backgrounds.
• Indoor Scenes (40 videos): Long trajectories filmed in indoor environments, including transitions between indoor and outdoor. These challenging sequences contain textureless regions, dynamic elements, and challenging lighting.
• Outdoor Scenes (40 videos): Diverse dynamic environments filmed in unique outdoor locations. These include different weather and lighting conditions such as full sunlight and nighttime.

File Structure

Inside each of the scanning/, indoors/, and outdoors/ folders, the data is further divided into subfolders:

• part_1/, part_2/, ..., part_N/: Each part contains one or more video sequences captured with the same rotation of the 360° user camera. Intrinsic parameters for the 360 camera view are provided per part.

Additional Downloads

• Stereo & IMU Data: For some video, we provide synchronized stereo and inertial data. Each download contains the following structure:
  • LEFT/ – Folder containing left camera images.
  • RIGHT/ – Folder containing right camera images.
  • effective_fps.txt – A text file containing the effective frames per second.
  • stereo_transformation.npy – A NumPy file with the transformation matrix between the stereo cameras.
  • relative_pose_gt_to_zed.npy – A NumPy file containing the 4x4 transformation matrix representing the relative pose between the ground truth view and the ZED left camera.
  • relative_pose_zed_to_imu.npy – A NumPy file containing the 4x4 transformation matrix representing the relative pose between the imu sensor and the ZED left camera.
  • imu – A file containing inertial measurement unit (IMU) data for the specific video.

  - stereo-val.zip
  - stereo-test.zip

NVS Benchmark

We also propose a challenging benchmark for Novel View Synthesis (NVS) based on six sequences selected from our Princeton365 dataset. These scenes involve complete 360° camera rotations around reflective or non-Lambertian objects—ideal for evaluating methods that aim to reconstruct challenging materials under diverse lighting.

The NVS benchmark data and evaluation trajectories can be downloaded here:
NVS Benchmark Sequences

How to Unzip

Our datasets are provided as multi-volume .7z archives. To extract them, you will need the p7zip command-line tool.

1. Install 7-Zip: If you don't have it, install it using conda:

   conda install -c conda-forge p7zip

2. Download the Folder: Download the entire folder containing all parts of an archive (e.g., my_dataset.zip.001, my_dataset.zip.002, etc.). Ensure all parts are in the same directory on your machine.
3. Extract the First Part: Navigate into the folder you just downloaded, and run the extract command on the first file only. 7-Zip will automatically find and combine the other parts.

   7z x my_dataset.zip.001

   Note on Extra Data: If the folder also contains an extra_data.zip.001 archive, you must unzip it as well:

   7z x extra_data.zip.001

   After extracting, merge the contents of the extra_data folder into the main dataset folder using rsync. This command will copy the new files without overwriting existing ones. For example:

   rsync -av --ignore-existing /path/to/extra_data/ /path/to/main_dataset/

For any questions or assistance, please contact us at: princeton365.benchmark@gmail.com