InspireFace v1.1.12

In the current version, we support Rockchip RV1106 and RK356x devices (RV1103 support is possible but unverified). We have implemented RGA hardware acceleration for image processing on devices with RKNPU2 support, significantly improving processing speed. Additionally, we provide an optimized Android SDK with JNI integration and a simple demo. Corresponding resource files are available for these platforms.

• Open the parameter interface of some trackers.
• Add a similarity conversion tool.
• Update t3 series model: Formalizes the structure of the description file.

InspireFace v1.1.11

In the current version, we support Rockchip RV1106 and RK356x devices (RV1103 support is possible but unverified). We have implemented RGA hardware acceleration for image processing on devices with RKNPU2 support, significantly improving processing speed. Additionally, we provide an optimized Android SDK with JNI integration and a simple demo. Corresponding resource files are available for these platforms.

• Fixed some bugs running on RV1106/1103 and RK356x devices.
• Fixed a bug where the database persistence save path was invalid.

InspireFace v1.1.10

In the current version, we support Rockchip RV1106 and RK356x devices (RV1103 support is possible but unverified). We have implemented RGA hardware acceleration for image processing on devices with RKNPU2 support, significantly improving processing speed. Additionally, we provide an optimized Android SDK with JNI integration and a simple demo. Corresponding resource files are available for these platforms.

• Add support for NPU acceleration inference on RK356X platforms
• Make keypoint detection optional by default, auto-enabling only in tracking mode
• Fix face bounding box displacement issue in detection mode

InspireFace v1.1.9

In the current version, we have adapted and tested for Rockchip's RV1106 device and provided corresponding resource files (it may support RV1103, but we haven't verified it with actual devices). We believe we will soon adapt to other Rockchip device models, such as RV356x and RV3588. Meanwhile, we are implementing RGA image hardware acceleration processing adaptation for devices supporting RKNPU2, which has improved the image processing speed on RK devices. We have improved the Android SDK based on Java Native Interface (JNI), optimized its size, and provided a simple demo.

• Add NPU inference support for Rockchip rv1106;
• Add new models to the Model Zoo;
• Fixed some bugs that were causing crashes;
• Added RGA acceleration support for some image processing interfaces on Rockchip devices with RKNPU2;
• Added a simple Android example demo;
• Added library support for Linux and macOS on x86 and arm64 platforms in PyPI, enabling rapid deployment;
• Release of precompiled libraries for macOS.

InspireFace v1.1.8

Adapted the optional image processing engines within the SDK, providing a more lightweight InspireCV while retaining OpenCV from previous versions; The SDK has been streamlined by removing some third-party dependencies, making it more lightweight.

1. Added and set as default a more lightweight image processing engine;
2. Modified the vector management engine of the Feature-Hub module to sqlite-vec, improving search efficiency;
3. Replaced internal data structures with generic abstract classes to ensure generalization;
4. Achieved overall SDK size reduction, resulting in a more lightweight library through compilation;
5. Implemented PyPI package management and enhanced the implementation of Python native interfaces.

InspireFace Model Zoo

Model Zoo

We provide resource packages (including models, configuration files, etc.) that are supported by InspireFace across different platforms. For more information, please refer to the README.

Different target platforms will use different compilation options and require different resource files. Generally, Pikachu and Megatron are universal models that work across all platforms, while the Gundam series needs to be selected based on the NPU support of your target device that you're compiling for.

Version: t3 Series

Normalize the version information and add other parameter structures besides the model configuration.

1. Pikachu

Lightweight model package for mobile devices using CPU inference.

2. Megatron

More professional face recognition model, suitable for mobile, server and other CPU or GPU reasoning scenarios.

3. Gundam_RV1109

Rockchip specific model for NPU inference of RV1109/RV1126 devices.

4. Gundam_RV1106

Rockchip specific model for NPU inference of RV1103/RV1106 devices.

5. Gundam_RK356X

Rockchip specific model for NPU inference of RK3566/RK3568 devices.

InspireFace v1.1.7

1. Fixed some feature hub persistence bugs;
2. Add test cases for the feature hub.

InspireFace v1.1.6

1. Added face action detection.
2. Fix some bugs and add test cases.
3. Add global resource statistics monitoring to prevent memory leaks.

InspireFace v1.1.5

1. Added facial action detection capabilities to the facial interaction module.
2. Fixed several bugs in the facial tracker.

InspireFace v1.1.4

1. Fix bugs in the Ctypes calls to native interfaces in Python.
2. Implement eye state prediction functionality in the facial interaction module.
3. Reorganize the directory structure of resource files.