> For the complete documentation index, see [llms.txt](https://doc.celantur.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.celantur.com/sdk/release-notes.md).

# Release Notes

## 2026

### Version 1.5.2

Released on August 7th, 2026.

* Support for NVIDIA Blackwell architecture (compute capability `10.0`).
* Now also supports Ubuntu 24.04 as base operating system.
* A new improved model with two different input sizes.

#### Dependencies 

* OpenCV v4.11.0
* TensorRT v10.15.1
* ONNX v1.20.1
* OpenVino v2024.6.0

### Version 1.5.1

Released on June 3rd, 2026.

* **Full GPU inference:** Use data in GPU memory as input and output for inference instead of copying from CPU memory. See [example](https://github.com/celantur/SDKExample/blob/main/gpu_inference.cpp).
* Run models with different input sizes. A **smaller, faster model** (with 640x640 input size) is now available upon request.
* **Improved default model** (segmentatio-omnium v10) trained with additional indoor, outdoor and dashcam images.

#### Dependencies

Same as for [#version-1.1.0](#version-1.1.0 "mention").

### Version 1.4.0

Released on March 19th, 2026.

* Add object tracking.
* Separate detection from blurring. See [example](https://github.com/celantur/SDKExample/blob/v1.4.0/detect.cpp).

#### Dependencies

Same as for [#version-1.1.0](#version-1.1.0 "mention").

### Version 1.3.0

Released on January 27th, 2026.

* Remove dependency on protobuf.
* Available with new model (segmentatio-omnium v8).

#### Dependencies

Same as for [#version-1.1.0](#version-1.1.0 "mention").

## 2025

### Version 1.2.0

Released on October 21st, 2025.

* Beautify blur for bounding boxes.
* SKD supports smaller models.
* TensorRT support.
* Fix bug where tiling and segmentation mask creation crashes the application.

#### Dependencies

Same as for [#version-1.1.0](#version-1.1.0 "mention")

### Version 1.1.0

First public release of Celantur SKD on July 18th, 2025.

* C++ library.
* Inference with ONNX and OpenVINO.
* Blurring of faces, license plates, persons and vehicles.
* Blurring using segmentation mask or bounding boxes.
* Detection metadata as structured data or JSON string.
* Custom tiling.
* Setting detection thresholds.
* Read and save JPEG images.
* Multithreading.
* Video processing plugin.
  * H.254 en- and decoding.
* Deliverable as Docker image or Debian libraries.

#### Dependencies

* Boost v1.84.0
* ONNX v1.20.1
* OpenCV v4.7.0
* TensorRT v10.0.1.6
* OpenVino v2024.6.0
* Optional: libjpeg-turbo v3.1.0