[ Home : Programs | libCVD | Hardware hacks | Publications | Teaching | TooN | Research ]

FAST Corner Detection -- Edward Rosten

FAST corner detection

Try FAST Today!

If you use FAST in published academic work then please cite both of the following papers:
  1. Fusing points and lines for high performance tracking.
  2. Machine learning for high-speed corner detection.
FAST-ER is now accepted for publication:

Any figures ma be reporduced with appropriate citations. For convenience, the FAST corner figure is available in a variety of formats here.

If you want to use FAST, it is available in a variety of forms below:

Questions about FAST

If you have any questions, try the FAQ, or ask a question about FAST in the forum.

Precompiled FAST binaries

FAST is available precompiled for a wide variety of platforms: The FAST binary accepts an image as input and output image with corners drawn on or a list of corner locations. The threshold, number of points in the detector (9 to 12) and nonmaximal suppression can be selected. See the unix README or Windows README for more details.


The windows executable has problems dealing widths which are not a multiple of 4. This bug does not affect the unix exectuables or the libraries.

FAST Source code

Python Source Code

Standalone C soure code

Note: this code is stable, not abandoned. The algorithm is fixed and this is the basic reference implementation. It will only change if bugs are found.



If you have the vision toolbox, then it is built into MATLAB and you don't need to download anything.

Source code for the executables

FAST in libCVD

Build your own FAST detector

There are several options. The easiest is to use an learned detector, but output it in a new language. The FAST-ER code below contains pre-made trees in an easy to use format and a number of code generators. If you wish to build your own FAST detector (e.g. trained on your own data, targeting another language, or using some new optimizations), then the FAST-ER code provides programs for training new FAST-N detectors as well as FAST-ER detectors.

Or, you can use the original version, from the 2006 paper:

Compare your detector to FAST

If you wish to compare your detector to FAST, then there is a set of registered images available for download. These were used for testing FAST in Machine learning for high-speed corner detection.. Source code for performing the comparisons is available in the FAST-ER distribution below.

FAST-ER: Enhanced repeatability

Source code for FAST-ER is now available: This code is useful if you want to: Documentation is available here.

People who use FAST

If you use FAST in your project, let me know and I'll put a link to it right here.


I would like to additionally thank Gabriel Brostow for help preparing this web page and the test executables.

Updated February 25th 2018, 05:59