Releases: mooch443/trex
v1.1.9
Every operating system now has the same installation command again!
# for Windows, Linux and macOS
conda create -n track -c trexing trex
- removed conda-forge dependency for linux (finally), which makes installation easier
- add visual_identification_version to allow switching to previous network architectures (loading old weights)
- add support for physical tags (square rectangles attached to individuals) both for segmentation as well as recognition (see
tags_enable
in the docs) - various GUI and user-experience updates
- increased processing speed for some platforms in TGrabs by up to 25%
- .settings.backup now goes into a different folder
- fixed some export bugs
- add blob_split_algorithm, which switches between thresholding and a… by @mooch443 in #161
- update conda-forge removed by @mooch443 in #173
v1.1.8_3
This is just a bug fix release (hopefully).
There was a bug in TRex that could lead to frame_segments being broken in exported data.
There was also a potential crash when loading files with -analysis_range [] set.
Categorization scaling was off on some platforms - hopefully now it won't be anymore.
What's Changed
- Merge back into locusts by @mooch443 in #148
- An issue with video lengths by @mooch443 in #147
- Some fixes from upstream by @mooch443 in #151
Full Changelog: v1.1.8_1...v1.1.8_3
v1.1.8_1
v1.1.7 had to be retracted and is now replaced with this version, including some important hot-fixes.
These include fixes for tracklet image export, which was broken in some versions before. Loading times have been improved, especially for videos/results with recognition data.
It follows the changelog from v1.1.7:
This release may have the occasional debug print-out glitch (already fixing those), but many other bugs, that may have never manifested before, have been fixed. Especially TGrabs has received significant improvements on Windows (i.e. finally sporting functioning CTRL+C behavior and usign dark-mode if available / Win11), as well as in being more consistent with how many frames are actually saved. Main focus have been code modernizing, efficiency improvements, and bug fixes - with only a few changes regarding the GUI.
This release also updates the compiler used on all systems (in conda), which lead to problems on Linux (highest version available is gcc 9.3). Consequentially, the install command on Linux has changed to:
# Linux
conda create -n tracking -c defaults -c conda-forge -c trexing trex
All other systems still use the shorter version:
# Other OS's
conda create -n tracking -c trexing trex
Minor issues resolved include #92 and probably #88.
Full Changelog: v1.1.6...v1.1.8
What's Changed
- Update README.md by @mooch443 in #79
- updating windows meta/tensorflow deps by @mooch443 in #81
- Merging changes for v1.1.7 by @mooch443 in #99
- Testing compilation issues by @mooch443 in #109
- UX Fixes by @mooch443 in #110
- training dialog / auto correct by @mooch443 in #120
- Fixes for the auto correct dialog + loading times by @mooch443 in #121
v1.1.6
v1.1.5
This update includes some bug fixes for crashes in relation to GUI, as well as restoring full functionality of other features during conversion (such as image_invert). With 1.1.5, TRex also allows users to install the newest Tensorflow releases for macOS (with updated instructions).
What's Changed
- Bug fixes by @mooch443 in #73
- Fixing issues (with image_invert, etc.) by @mooch443 in #76
- meta.yaml update for newest macos tensorflow by @mooch443 in #77
Full Changelog: v1.1.4...v1.1.5
v1.1.4
v1.1.3
This update adds a new feature "Categorization", which allows users to classify individuals based on their appearance - independently of their identities. Choose one from a range of animated pictures and assign a category, until the neural network yields satisfactory results!
v1.1.3 also contains a series of bug fixes and improvements.
- visual and readability improvements
- other GUI improvements, such as DPI-related display issues
- changed
blob_id
hash (see below) - fixing compilation issues with new libpng
- happy mode
One change that might affect analysis of older data is how blob_id
is calculated now: Instead of calculating the center of objects and using this coordinate to calculate a hash (the blob_id
), the new hash is based on the top-left coordinate + the bottom-right x coordinate + the number of horizontal lines in an object. This is both much faster to calculate (no iterating through all lines) and more distinct (even slight changes in object size will be noticed and objects can theoretically occupy the same space and still be distinct).
v1.1.2
This release contains many misc bug fixes and improvements, including
- meta strings for vectors of objects were not produced correctly, leading to gui bugs
- GCC10 support
- multiple scaling bugs fixed (including OS-specific ones and regarding DPI)
- performance improvements in CPULabeling, especially with many objects
- fix issues that could lead to TGrabs (and sometimes TRex) crashing. Especially on Windows.
- fixed CSV export on windows sometimes exporting double-line breaks
- fixed AveragingAccumulator (preview in VideoOpener dialog)
- fixed a bug that consistently produced outliers when individuals had been lost for at least a frame
- fixed tensorflow initialization bug
v1.1.1: GUI updates, etc.
Aside from fixing a couple of GUI issues, this update should improve the overall feel and functionality of TRex (and TGrabs). I also added the parameter httpd_accepted_ip
for versions of TRex compiled with HTTPD support, improving security by only permitting one specific IP address to control the program.
This also (likely) fixes an issue, common on macOS, which made TRex crash before opening TGrabs to convert videos, and adds better documentation throughout.
Updating to tensorflow 2.x, and Support for Apple M1 (arm64)
Installation works as usual:
conda create -n tracking -c main -c conda-forge -c trexing trex # linux
conda create -n tracking -c trexing trex # other
Network architecture has been changed/layers with the same function replace old ones. This means that loading old network weights might not work.
When using old-layout network weights, simply create a conda environment with an old TRex:
conda create -n tracking -c main -c conda-forge -c trexing trex==1.0.10 # linux
conda create -n tracking -c trexing trex==1.0.10 # other
I have not tested upgrading an old conda environment to this version, so I recommend creating a new environment.
See #9 (comment) for details.