## [2.0.1] - July 12, 2022
  * fixed multiple bugs that were causing server crash
    - handle requests for empty image asset names (e.g. GET image or
    GET image/)
    - registration process of im1 and im2 transforms im2 to the im1
      coordinate space, this transformation was wrongly using the size
      of im2 as target size when it should be using size of im1 as the
      target size, this crashed the server when crop bounding box fell
      beyond image size
    - show error message when uploaded image exceeds size limit;
    removed debug messages
    - show messages when external search returns no result
    - thread queue now wais for all threads instead of waiting for
    only N-1 threads

  * improved documentation
    - doc/HTTP-REST-API.md :
      - POST _search_using_features requires image features to be sent
      as request body
      - added doc about performing external image search using HTTP
      REST API (thank you RKD team for the question)
    - doc/Users-of-VISE.md : lists research projects and commercial
    products that use VISE
    - doc/Visual-Vocabulary.md : added details about generic visual
    vocabulary that ships with VISE
    - doc/FAQ.md :
      - use hamm64 generic visual vocabulary as hamm32 slows search on
      large datasets; added link to generic visual vocabulary page
      - added details about using VISE in Windows server.
      - define the metadata fields that gets shown in the user interface
    - doc/Testing-VISE.md : added doc for VISE tests
    - doc/User-Guide.md : added User Guide

  * improved build script
    - install libjpeg and libpng libraries which are required by
    ImageMagick (thank you @mchesterkadwell)
  * user uploaded images are resized before performing "upload and
  search"
  * fixed "undefined" region for image search using full image
  * rectified errors pointed by static analysis (cppcheck)
    - close files before premature return
    - typo in variable name
  * removed leading space character in each line of conf.txt file
  entries in User-Guide (thank you leander-de-graaf)
  * fixed Windows CI/CD build by adding double quotation to msbuild
  command parameter
  * set visualization must show original image (not thumbnail) as
  region dimensions are only correct in original image space
  * turn on HTTP request logging only in debug builds
  * added visual grouping feature (available only through vise-cli
  command)
  * fixed indexing time message to properly show hour field > 99
  * fixed typo in project CSS display:flex (thank you
  @mchesterkadwell)
  * improved some user interface elements based on feedback from users

## [2.0.0] - June 14, 2021
  * removed dependency on Python:
    - clustering of visual features done using vlfeat's kdtree implementation (added dependency on vlfeat)
  * Simplified user interface which is now based solely on HTML/CSS/Javascript
    - frontend is based solely on HTML/CSS/Javascript (previously, VISE used a python cherrypy based server for generating frontend)
    - many work load transferred from backend server to user's web browser
    - VISE server only generates basic HTML pages containing dynamic data stored as Javascript variable
    - web browser renders the page based on dynamic data contained in Javascript variable
    - VISE servers sends homography matrix and web browser uses Javascript and HTML canvas to generate registered images and required visualisations
  * upgraded code to support compilation in Linux (gcc), Windows (Visual C++) and MacOS (Clang)
  * setup CI/CD pipeline (build and test stages) for Windows 10
    - automatically builds using Visual C++ compiler on Windows 10 platform
    - integration test based on Oxford Buildings 100 images subset and its ground truth derived from the full Oxford Buildings 5k dataset
    - uses NSIS to package Windows 10 64 bit installer
  * images get resized based on user defined max. size (defined in project configuration file)
  * indexing process log saved to a file for reporting
  * a basic GUI application to launch VISE server in Windows 10
  * faster indexing of projects using generic visual vocabulary
    - VISE ships with a generic visual vocabulary computed using 53629 images
    - this speeds up the process of indexing (at the loss of some search accuracy) of new projects
  * rewrote VISE http server
    - all functionality of VISE software is now exposed using a REST HTTP API
    - multiple worker threads handle HTTP requests
  * following software features that have been implemented but not yet fully tested have been queued for future release:
    - visual grouping of images
    - metadata import and search

## [2.0.0-internal-releases] - 2018, 2019, 2020
  * VISE code was adapted and extended to address the requirements of several internal and collaborative projects at VGG between 2018 and 2020.

## [1.0.3] - September 28, 2017
  * international characters (utf-8) are now allowed in image filenames (issue 15)
  * uppercase image filename extensions (PNG, JPG) are now allowed (issue 17)

## [1.0.2] - September 07, 2017
  * docker container for Linux, MacOS and Windows

## [1.0.1] - July 05, 2017
  * fixed issues with load, start, stop and uninstall scripts
  * debugging log saved to $HOME/vgg/vise/log/training.log
  * user images are resized *only* if > 1024px
  * added Refresh button to indexing page to deal with unresponsive pages (temporary fix)

## [1.0.0] - June 15, 2017
  * a web based user interface to index new images

## [0.0.2] - March 16, 2017
  * updated source code and cmake files to allow compilation in Mac OS
  * mostly used by @Giles Bergel (Seebibyte project ambassador) to demonstrate potential users

## [0.0.1] - September 25, 2014
  * the original relja_retrival codebase was released internally within the Visual Geometry Group (VGG)
  * developed by Relja Arandjelovic during his DPhil / Postdoc at the Visual Geometry Group, Department of Engineering Science, University of Oxford.
