deeptrack_dataset

Python3 code to load and evaluate 6DOF tracking dataset from Deep 6 DoF Tracking

Dependencies

• PIL
• vispy
• pyopengl
• opencv

Download datasets

Script to download/extract the full dataset from the servers. Contains option to skip downloads and to set output paths. The default path is ./data.

    python download.py -o ./data

• 3dmodels contains each object models
  • geometry.ply : vertex (meter scale), color, texture information
  • ao.ply : per vertex ambiant occlusion intensity
  • object.bmp : texture
• raw_training contains training data for each objects.
  • x.png : rgb frame id x
  • xd.png : depth frame id x
  • camera.json : camera intrinsics
  • viewpoints.json : camera extrinsics (see show sequence example for an example to load them
• sequences contains test data for each objects.

Show sequence

Script to visualize the dataset (sequence only). Serve as an example on how the dataset is loaded in memory. The option --save will save the ground truth data in the format used for evaluation.

    python show_sequence --dataset ./data --object dragon

To load result instead of ground truth, use --load option:

    python show_sequence --dataset ./data --load ./sequence_dragon.npy

TODO

• Add support for csv file
• evaluation code