Skip to content

Changelog

All notable changes to ecos_core are documented here, grouped by release version (see Versions for download links).

v2.0.0

  • Breaking: removed unused/deprecated modules: SAHI, efficienet-pytorch, efficientdet-pytorch, gan (DCGAN, Progressive-GAN), head_pose, kanezaki, opencv_holes, resnet_pytorch, resnet_regression, shape_guided3D, solov2, tam, vitpose, yolo_nas.
  • Breaking: pip install ecos_core now only packages indad, yolo8, paddle_ocr, groundingDino (plus base_model/common). SAM, langSAM, autodistill, fastsam and loca are no longer part of the installable package — they remain in the repository for investigation use, with their dependencies covered by the repo's requirements.txt instead of setup.py.
  • Added paddle_ocr module (PaddleOCR wrapper).
  • Removed the segment_anything post-install step (only needed by SAM/langSAM, no longer shipped by default).
  • Cleaned up API/application docs pages that referenced the removed modules.

v1.4.1

  • Added loca (zero/one-shot object counting model).
  • Added vision_benchmarker.
  • Added autodistill, including a SAHI + autodistillation workflow.
  • Fixed supervision version pin.

v1.4.0

  • Added yolo_nas.
  • Added shape_guided3D (shape-guided 3D anomaly detection).
  • Added hyper_optimizer for indad.
  • Improved fastsam (dynamic term, load speed) and indad PatchCore (clustering memory bank).
  • Added a background-removal tool.

v1.3.0

  • Added groundingDino.
  • Added a SAM + SOLOv2 water-level measurement pipeline.
  • Fixed indad backward-compatibility issues.

v1.2.1

  • Added langSAM.
  • Added vitpose.
  • Added SA-PatchCore and cross-validation threshold selection for indad.
  • Fixed indad compatibility and setup.py packaging issues.

v1.1.0

  • Added SAM.
  • Added solov2.
  • Added the finalized yolo8 package.
  • Improved indad (PaDiM/SPADE support, auto-augmentation).

v1.0.0

  • Initial packaged release: indad anomaly detection (PatchCore).