Releases: timmh/distance-estimation
Release v1.6.3
This release fixes an issue on Windows machines which is caused by sys.stdout
and sys.stderr
being None
when running the GUI without attached console.
Release v1.6.2
This release implements the original PyTorch implementation of DPT, in order to be able to reproduce results in Haucke et al. 2022 as closely as possible (though still relying on MegaDetector v5 instead of v4).
Release v1.6.1
This release makes the downloading of model weights more robust by checking MD5 checksums and rejecting the files were checksums don't match.
Release v1.6.0
Make model fitting deterministic and more reliable by ignoring background pixels at or above the maximum distance.
Release v1.5.1
This release fixes overly lenient parsing of calibration distances from filenames, which could lead to filenames of calibration images not representing distances being interpreted as such.
Release v1.5.0
This release adds an option to mask animals during calibration, which should improve results for very close animals (especially in conjunction with SAM).
Release v1.4.3
This release allows non-integer calibration distances is more lenient when parsing calibration distances from filenames
Release v1.4.2
This release fixes the detection angle computation for good by switching from xywh to correct xyxy bouding box format.
Release v1.4.1
This release fixes a bug in the computation of detection angles that created a bias towards larger angles.
Release v1.4.0
This release add support for metric depth estimation without reference images using Depth Anything