The Learned Arrangements of
Three Patch Codes (LATCH) project

Visualization of the LATCH descriptor
Figure 1. Visualization of the LATCH descriptor. Given an image patch centered around a keypoint, LATCH compares the intensity of three pixel patches in order to produce a single bit in the final binary string representing the patch. Example triplets are drawn over the patch in green and blue.

What is LATCH?

LATCH is a fast and compact binary descriptor used to represent local image regions. LATCH was shown to obtain better performance on standard benchmarks than other pure binary descriptors, sometimes matching the accuracy of far larger floating point representations such as SIFT. Its extraction time is only slightly longer than other binary descriptors and far faster than floating point representations. See the LATCH paper below for further details.

LATCH is now part of the OpenCV xfeatures2d library, since version 3.0.

LATCH won the CVPR 2015, OpenCV State of the Art Vision Challenge, in the Image Registration category.

A video of the talk given by Gil at WACV'16, describing LATCH:

What is CLATCH?

Pronounced clutch, it is a CUDA port of the LATCH descriptor to the GPU. It is an ongoing project maintained by Christopher Parker, Matthew Daiter and Kareem Omar. CLATCH is an extremely fast local descriptor. In fact, to our knowledge, it is presently the fastest available local binary descriptor. See the CLATCH paper for runtime comparisons.

What are LATCH/CLATCH used for?

They can be used as alternatives to other local image descriptors, such as SIFT, SURF and ORB. In particular, CLATCH was integrated into the OpenMVG Structure from Motion (SfM) library and used to produce extremely fast, detailed 3D reconstructions from high resolution photos.

Relevant references:

  • Christopher Parker, Matthew Daiter, Kareem Omar, Gil Levi and Tal Hassner, The CUDA LATCH Binary Descriptor: Because Sometimes Faster Means Better, Workshop on Local Features: State of the art, open problems and performance evaluation, at the European Conference on Computer Vision (ECCV), Amsterdam, The Netherlands, Oct. 2016 (PDF, BibTex)

  • Gil Levi and Tal Hassner, LATCH: Learned Arrangements of Three Patch Codes, IEEE Winter Conference on Applications of Computer Vision (WACV), Lake Placid, NY, USA, March, 2016 (PDF, BibTex)

Code and resources

  • (CPU) OpenCV LATCH: The original LATCH OpenCV implementation. It is part of the xfeatures2d library. The full, merged pull request is also available. Example usage of this version is available in LATCH_match.cpp.

  • (CPU) LATCH: A new more efficient reimplementation of LATCH. Our tests show it to run ~29 times faster than the original OpenCV implementation.

  • (CPU) ULATCH: A restricted, upright only version of LATCH, (non rotationally invariant) descriptors. It is the fastest CPU implementation of LATCH making it well suited for feature tracking applications on the CPU.

  • (CUDA-GPU) CLATCH: Code for CLATCH, the CUDA LATCH GPU port.

  • (CUDA-GPU) UCLATCH: A restricted, Upright CLATCH, which reaches even faster speeds than CLATCH on the GPU by restricting descriptors to upright only.

  • A branch of the OpenMVG, Structure from Motion library, which includes CLATCH to provide high speed 3D multi-view reconstructions. Also new to this branch is a GPU based, brute force Hamming distance matcher for binary descriptors.
If you find any of these resources useful, please be sure to cite the relevant papers above.

The people involved in this project

In alphabetical order:

LATCH Copyright 2015, Gil Levi and Tal Hassner

CLATCH copyright 2016, Christopher Parker, Matthew Daiter, Kareem Omar, Gil Levi and Tal Hassner

The SOFTWARE provided in this page is provided "as is", without any guarantee made as to its suitability or fitness for any particular use. It may contain bugs, so use of this tool is at your own risk. We take no responsibility for any damage of any sort that may unintentionally be caused through its use.

Last update 14th of Sept, 2016