Datasets:

ai4ce

/

EgoPAT3Dv2

like 0

Follow

ai4ce@nyu 49

EgoPAT3Dv2

Dataset introduction

There are 11 scenes contained in the EgoPAT3Dv2 dataset, corresponding to folders 1 through 11. Each scene folder contains 2 to 6 video folders, and each video folder contains an RGB folder, a depth folder, a point cloud folder and a transformation matrices folder. (Please ignore other folders or files inside the zip file.) The annotations (ground truth) and transformation matrices (the same as the transformation matrices above) are included in the annotation_transformation.hdf5 file. We use HDF5 to organize the dataset in the experiment, and the dataloader in the GitHub repo is also written correspondingly.

Dataset folder hierarchy

Dataset/
    ├── 1	# scene 1
        ├── 1.1.zip -> 1.1		# video 1 in scene 1
            ├── d2rgb						# depth files
            ├── color						# rgb files
            ├── pointcloud			# point cloud files
            └── transformation	# transformation matrices
        ├── 1.2.zip -> 1.2 	# share the same structure as 1.1
        ├── ...
        └── 1.4.zip -> 1.4
    ├── 2/ # all scene/video directories share the same structure as above
        └── ...
    .
    .
    .
    
    └── 11

Construct HDF5 dataset file

Since 50GB is the hard limit for single file size in huggingface, please use make_RGB_dataset.py to construct the HDF5 file on your own.

1. Download all zipped files. Unzip them and keep RGB("color" in the folder) folder in each video folder only.
2. Run make_RGB_dataset.py after step 1.