deeptracking

Torch + python implementation of Deep 6-DOF Tracking

Note that a docker file is available and a docker image is upload to the dockerhub but it can not generate a window for generating data so it can only used for training.

The code to parse a ply file is not provided, this framework needs an implementation of plyparser.py

TODO: update with the more recent pytorch framework...

The datasets and 3D models from the paper can be found Here.

Dependences installation

Please refer the odt files for more details about this project (installation ,dataset ,file change etc.) There should have everything you need for the project including setup the environment , make up json , download dataset and some issues you might meet when running the project(there should be a plyparser issue in my develop repo which is what I am dealing with it.Just as a suggestion ,you might need try some different version and diffrent installation ways of plyfile to have a test of it .Making plyparser work is just important than everything as other thing I have shared the details.Tested in ubuntu14.4.5,python3.4 as default.)BTW,you can also refer the wiki to install everything and have a glance of the whole project:https://github.com/Myzhencai/deeptracking/wiki

Warnings

This project use python3 and Gpu with cuda cause there have the cnn etc.And I just installed the pytorch from source after I installed Hugh Perkins's pytorch,I do not know if it is neccessary so I did not mention it in the odt files.

Generating data

Generating synthetic data

Run:

python generate_synthetic_data.py config_file.json

dependencies

• cv2
• tqdm
• pyOpenGL
• glfw
• numpngw

configuration

see this example file

Generating real data

Run:

python generate_real_data.py config_file.json

dependencies

• cv2
• tqdm
• pyOpenGL
• glfw
• numpngw

configuration

see this example file

Train

Run:

python train.py config_file.json

dependencies

• Hugh Perkins's pytorch
• scipy, skimage, numpy
• tqdm
• numpngw
• slackclient (could be removed)

configuration

see this example file

Test

Sensor

Will run the tracker with a sensor (kinect 2)

python test_sensor.py config_file.json

Sequence

Will run the tracker on a sequence (folder) and save the error in a file

python test_sequence.py config_file.json

dependencies

• cv2
• Hugh Perkins's pytorch
• pyOpenGL
• glfw
• numpngw
• pyfreenect2 (for Kinect2 sensor)

configuration

see this example file