b51 / FeatureTracker

Features detection, match and track module. Supported features: orb and superpoints
MIT License
30 stars 6 forks source link

FeatureTracker

What this repository for

Detect, match and track different type features

Current type of feture points

Usage

PreInstall

install libpytorch: Install from source with Anaconda Set libpytorch cmake path at CMakeLists.txt

Dependence Info

pytorch: 1.1.0

cuda/cudatoolkit: 9.0

Build

$ cd FeatureTracker
$ mkdir build
$ cd build && cmake ..
$ make -j4

Run ORB feature points

$ ./FeatureTracker --image_dir /path/to/images_dir --type orb [--image_suffix .jpg]

SuperPoint feature points

Download superpoint model
$ cd utils
$ wget https://github.com/MagicLeapResearch/SuperPointPretrainedNetwork/raw/master/superpoint_v1.pth
$ python pytorch_model_convert.py superpoint_v1.pth
$ cd ../build
$ ./FeatureTracker --image_dir /path/to/image_dir --model_path ../utils/super_point_scripted_model.pt --type sp [--image_suffix .jpg --H 480 --W 640]

Kitti:

ORB Tracker

SuperPoint Tracker

Resized image to 414, 125 (width/3, height/3)

Original size image