readme.md

Preprocessing Tensor4D Dataset

Example: yxd-6views

Once you have extracted the data, the data directory structure should be as follows:

yxd-6views
├── 20114279.hevc
├── 20114281.hevc
├── 22053924.hevc
├── 22053927.hevc
├── 22138903.hevc
├── 22139915.hevc
└── calibration_full.json

There are three steps to process the data:

1. Transform the hevc format videos to jpg images and save extrinsics and intrinsics.

   python preprocess.py --mode 'process' --data_dir /path/to/data/directory

2. Create the masks with any segmentation or matting methods. The required directory structure of one camera view should be as follows:

    20114279
    ├── color00000.jpg
    ├── color00001.jpg
    ├── ...
    ├── mask00001.jpg
    ├── mask00001.jpg
    ├── ...
   

3. Generate data with Tensor4D format.

   python preprocess.py --mode 'generate' --start_frame ${start_frame_index} --n_frames ${number_frames} --data_dir /path/to/data/directory/of/step1 --output_dir /path/to/output/directory

   --start_frame: Specifies the starting frames of the data used for converting to the Tensor4D format.

   n_frames: Specifies the number of frames required in each view.

   The final output directory structure:

   output_dir
   ├── cameras_sphere.npz
   ├── image
   └── mask