ailia-models
The collection of pre-trained, state-of-the-art models.
Setup
About ailia SDK
ailia SDK is a cross-platform high speed inference SDK. The ailia SDK provides a consistent C++ API on Windows, Mac, Linux, iOS, Android and Jetson. It supports Unity, Python and JNI for efficient AI implementation. The ailia SDK makes great use of the GPU via Vulkan and Metal to serve accelerated computing.
Install ailia SDK
- Download a free evaluation version of ailia SDK
- Unzip ailia SDK
- Find the location of Python site-packages directory
python -c "import site; print (site.getsitepackages())"
- Copy the ailia folder located in the python folder to site-packages
- Copy library files (dll or dylib or so) from the folder library to site-packages/ailia
- In the evaluation version, place the license file in the same folder as libailia.dll on Windows and in ~/Library/SHALO/ on Mac.
Install required libraries for Python
For Windows, Mac, Linux
pip install -r requirements.txt
For Jetson
sudo apt install python3-pip
sudo apt install python3-matplotlib
sudo apt install python3-scipy
pip3 install cython
pip3 install numpy
OpenCV for python3 is pre-installed on Jetson. You only need to run this command if you get a cv2 import error.
sudo apt install python3-opencv
Supported models
Image classification
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
vgg16 | Very Deep Convolutional Networks for Large-Scale Image Recognition | Keras | 1.1.0 and later |
googlenet | Going Deeper with Convolutions | Pytorch | 1.2.0 and later |
resnet50 | Deep Residual Learning for Image Recognition | Chainer | 1.2.0 and later |
inceptionv3 | Rethinking the Inception Architecture for Computer Vision | Pytorch | 1.2.0 and later |
mobilenetv2 | PyTorch Implemention of MobileNet V2 | Pytorch | 1.2.0 and later |
mobilenetv3 | PyTorch Implemention of MobileNet V3 | Pytorch | 1.2.1 and later |
partialconv | Partial Convolution Layer for Padding and Image Inpainting | Pytorch | 1.2.0 and later |
efficientnet | A PyTorch implementation of EfficientNet | Pytorch | 1.2.3 and later |
Image segmentation
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
deeplabv3 | Xception65 for backbone network of DeepLab v3+ | Chainer | 1.2.0 and later |
hrnet_segmentation | High-resolution networks (HRNets) for Semantic Segmentation | Pytorch | 1.2.1 and later |
hair_segmentation | hair segmentation in mobile device | Keras | 1.2.1 and later |
pspnet-hair-segmentation | pytorch-hair-segmentation | Pytorch | 1.2.2 and later |
U-2-Net | U^2-Net: Going Deeper with Nested U-Structure for Salient Object Detection | Pytorch | 1.2.2 and later |
Image manipulation
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
srresnet | Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network | Pytorch | 1.2.0 and later |
noise2noise | Learning Image Restoration without Clean Data | Pytorch | 1.2.0 and later |
dewarpnet | DewarpNet: Single-Image Document Unwarping With Stacked 3D and 2D Regression Networks | Pytorch | 1.2.1 and later |
illnet | Document Rectification and Illumination Correction using a Patch-based CNN | Pytorch | 1.2.2 and later |
adain | Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization | Pytorch | 1.2.1 and later |
deep-image-matting | Deep Image Matting | Keras | 1.2.3 and later |
Image to vector
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
illustration2vec | Illustration2Vec | Caffe | 1.2.2 and later |
Object detection
Deep Fashion
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
clothing-detection | Clothing-Detection | Pytorch | 1.2.1 and later |
Object tracking
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
deepsort | Deep Sort with PyTorch | Pytorch | 1.2.3 and later |
Action Recognition
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
mars | MARS: Motion-Augmented RGB Stream for Action Recognition | Pytorch | 1.2.4 beta and later |
Pose estimation
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
openpose | Code repo for realtime multi-person pose estimation in CVPR'17 (Oral) | Caffe | 1.2.1 and later |
lightweight-human-pose-estimation | Fast and accurate human pose estimation in PyTorch. Contains implementation of "Real-time 2D Multi-Person Pose Estimation on CPU: Lightweight OpenPose" paper. | Pytorch | 1.2.1 and later |
lightweight-human-pose-estimation-3d | Real-time 3D multi-person pose estimation demo in PyTorch. OpenVINO backend can be used for fast inference on CPU. | Pytorch | 1.2.1 and later |
3d-pose-basline | A simple baseline for 3d human pose estimation in tensorflow. Presented at ICCV 17. | Tensorflow | 1.2.3 and later |
pose_resnet | Simple Baselines for Human Pose Estimation and Tracking | Pytorch | 1.2.1 and later |
Gaze estimation
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
gazeml | A deep learning framework based on Tensorflow for the training of high performance gaze estimation | TensorFlow | 1.2.0 and later |
Face recognization
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
face_classification | Real-time face detection and emotion/gender classification | Keras | 1.1.0 and later |
vggface2 | VGGFace2 Dataset for Face Recognition | Caffe | 1.1.0 and later |
facial_feature | kaggle-facial-keypoints | Pytorch | 1.2.0 and later |
face_alignment | 2D and 3D Face alignment library build using pytorch | Pytorch | 1.2.1 and later |
blazeface | BlazeFace-PyTorch | Pytorch | 1.2.1 and later |
arcface | pytorch implement of arcface | Pytorch | 1.2.1 and later |
prnet | Joint 3D Face Reconstruction and Dense Alignment with Position Map Regression Network | TensorFlow | 1.2.2 and later |
Rotation estimation
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
rotnet | CNNs for predicting the rotation angle of an image to correct its orientation | Keras | 1.2.1 and later |
Crowd counting
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
crowdcount-cascaded-mtl | CNN-based Cascaded Multi-task Learning of High-level Prior and Density Estimation for Crowd Counting (Single Image Crowd Counting) | Pytorch | 1.2.1 and later |
Depth estimation
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
monodepth2 | Monocular depth estimation from a single image | Pytorch | 1.2.2 and later |
midas | Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer | Pytorch | 1.2.4 beta and later |
Text recognization
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
etl | Japanese Character Recognization | Keras | 1.1.0 and later |
Natural language processing
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
bert | pytorch-pretrained-bert | Pytorch | 1.2.2 and later |
Generative Adversarial Networks
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
pytorch-gan | Code repo for the Pytorch GAN Zoo project (used to train this model) | Pytorch | 1.2.4 beta and later |
council-GAN | Council-GAN | Pytorch | 1.2.4 beta and later |
Commercial Model
Name | Detail | Exported From | Supported Ailia Version |
---|---|---|---|
acculus-pose | Acculus, Inc. | Caffe | 1.2.3 and later |