Released ailia SDK 1.2.6

Introducing version 1.2.6 of ailia SDK, a cross-platform, GPU-enabled, fast AI inference framework.

Added the ability to omit prototxt when using ONNX.

Added the ability to omit the prototxt when using ONNX. Previously, ONNX was instantiated as follows.

net = ailia.Net(“a.onnx.prototxt”,”a.onnx”)

Since ailia SDK 1.2.6, prototxt can be omitted as follows.

net = ailia.Net(None,”a.onnx”)

For Unity, specify null.

model.OpenFile(null,”a.onnx”)

In the case of C/C++, simply omit ailiaOpenStreamFile.

Enhanced layer support

We have added 5D input support for Softmax, ReduceSum, and Gather, 1D input support for Deconvolution. We also added Metal support for Space2Depth, 3DConvolution, and Split on macOS. This will allow you to run a wider range of models faster.

Add run API to ailia Python API

In order to improve consistency with ONNX Runtime, a new run API has been added. In the case of the conventional ailia predict API, if the input is a Tensor, the output is also a Tensor, and if the input is an Array, the output is also an Array, but with the new run API, the output is always an Array.

PoseEstimator API has been optimized for speed

The pre-processing and post-processing code included in the ailia PoseEstimator API has been optimized for speed. In the RTX2080 environment, LightWeightHumanPoseEstimation runs in about 12ms, enabling real-time skeleton detection.

Support for new AI models

The following models compatible with 1.2.6 have been added to ailia MODELS. ailia SDK now supports more than 100 models, bringing the total to 110 models.

3dmpe posenet : 3D skeletal estimation

Quotation：https://github.com/mks0601/3DMPPE_POSENET_RELEASE

ailia-ai/ailia-models(from https://github.com/mks0601/3DMPPE_POSENET_RELEASE/tree/master/demo) Automatically downloads the onnx and prototxt…github.com

Efficient pose : 2D skeleton estimation

ailia-ai/ailia-models(Image from https://github.com/daniegr/EfficientPose/blob/master/utils/MPII.jpg) Model variant: RT Ailia input shape …github.com

Person reid baseline : Same person determination

Quotation：https://github.com/layumi/Person_reID_baseline_pytorch

ailia-ai/ailia-models(Image from http://188.138.127.15:81/Datasets/Market-1501-v15.09.15.zip) Shape : (batch, 3, height, width)…github.com

Source seperation : separation of sound sources

ailia-ai/ailia-modelsNoisy speech (audio file) Audio from creative commons youtube videos…github.com

Yet-Another-Anime-Segmenter : Segmentation of Anime style Images

ailia-ai/ailia-models(Image from https://unity-chan.com/download/index.php and licensed under © Unity Technologies Japan/UCL) ailia input…github.com

EAST : Extract text area

Quotation：https://github.com/argman/EAST

Style2paints : Auto Color

Quotation：https://github.com/lllyasviel/style2paints

Crnn.pytorch : OCR for English

ailia-ai/ailia-models(Image above is from https://github.com/meijieru/crnn.pytorch.) The output character will be printed. Automatically…github.com

deep-text-recognition-benchmark : English OCR

ailia-ai/ailia-modelsPretrained models for ailia SDK. Contribute to ailia-ai/ailia-models development by creating an account on GitHub.github.com

PointNet.pytorch : Segmentation of point clouds

Quotation：https://github.com/fxia22/pointnet.pytorch

ailia-ai/ailia-models(Image from http://web.stanford.edu/~ericyi/project_page/part_annotation/index.html) Segmentation model Input Shape …github.com

Pixel-Link : Text area extraction

ailia-ai/ailia-models(Image from https://rrc.cvc.uab.es/?ch=4&com=downloads) Shape : (height, width, 3) pixel_pos_scores shape : (1, 192…github.com

Pytorch-dc-tts : Speech Synthesis Model for English

ailia-ai/ailia-modelsA sentence which is defined as SENTENCE in pytorch-dc-tts.py. The Voice file is output as .wav which path is defined as…github.com

ailia Inc. has developed the ailia SDK, which enables cross-platform, GPU-based rapid inference. ailia Inc. provides a wide range of services from consulting, model creation, SDK provision of SDKs, development of AI-based applications and systems, to support Please feel free to contact us as we offer a total solution for.