Update v1.2.1; DShow lib, RGBDT Capture app, and more cleanups

This release introduces, apart from numerous fixups/cleanups in modules and algos:

• New platform utility functions for path manipulation and file preallocation;
• New sample projects showing edge/change detection/datasets module usage (with READMEs);
• An overhauled and more flexible version of the datasets input/output data mapping scheme;
• The DirectShow (DShow) base classes along with new utilities as a new 3rd party library project;
• Kinectv1/v2 SDK modules along with a standalone copy of some data structures as platform utils;
• An OpenCV-compatible DShow asynchronous device/frame grabber;
• A worker pool (thread pool) implementation for quick async task completion
• New async writers in the datasets module for off-thread packet compression/writing with queues;
• New CxxUtils implementations (semaphores and unlock_guard).

Also, the framework can now be built using 32bit toolchains, and the "install" target has been fixed on MSVC.

Update v1.2.0; edge detection w/ binary feature convolutions

This release adds non-max suppression utilities to the imgproc module and introduces an implementation of Canny's method based on binary feature convolutions via LBSPs. The associated CVPRW paper should soon be available on the lab's publication page.

Besides, note that this release also introduces drastic SIMD optimizations for LBSP feature computations in the features2d module. This should positively affect the processing speed of all background subtraction methods that use LBSP features; for example, LOBSTER's processing speed on my laptop improved by about 60%, to nearly 70 FPS.

Update v1.1.0; datasets module

This release overhauls the (late) Dataset(Eval)Utils.hpp/cpp core utilities and exports them to their own separate module.

The new datasets module can be used via a single header, datasets.hpp. A "Dataset" object is composed of several templated interface layers, each specialized to address a single task and provide a single evaluation protocol. Task enumerators identify how the data should be parsed and presented, while evaluation enumerators identify how results should be received and treated.

Default interfaces for image-based and video-based dataset parsing are available, and should allow on-the-fly dataset parsing and evaluation, given proper local folder structure. Some datasets which require more specific handling have their own implementation specialization headers, and can be used out-of-the-box:

• Foreground-background video segmentation & background subtraction:
  • ChangeDetection.net 2012
  • ChangeDetection.net 2014
  • Wallflower
  • PETS2006 (dataset#3)
• Boundary segmentation & edge detection:
  • BSDS500

For usage examples, see the apps/changedet and apps/edges projects.

Minor update v1.0.1; MSVC2015 fixups

This release fixes multiple issues which prevented the project from running out-of-the-box with MSVC2015. It also includes minor code structure/formatting cleanups.

LITIV Framework v1.0.0 stable

First stable release of the LITIV framework;

Contains improved versions of previously-released algos:

• LBSP feature descriptor/matcher w/ vectorization support
• LOBSTER (non-parallel impl and GLSL impl, both with improved LBSP lookup)
• SuBSENSE (non-parallel impl)
• PAWCS (non-parallel impl)
• VirtualPTZ

Also provides dataset parsing/evaluation utilities for:

• BSDS500
• CDnet 2012/2014

TODO's related to this release:

• Add LBSP lookup improvement to SuBSENSE and PAWCS
• Finish SuBSENSE+PAWCS GLSL impl