ailia-models-tflite

Quantized tflite models for ailia TFLite Runtime

About ailia TFLite Runtime

ailia TFLite Runtime is a TensorFlow Lite compatible inference engine. Written in C99, it supports inference in Non-OS and RTOS. It also supports high-speed inference using Intel MKL on a PC. In the Android environment, we provide a Unity Package, which also supports NPU inference using NNAPI.

Install

NEW - ailia TFLite Runtime can now be installed with "pip3 install ailia_tflite" !

Run the following command. The Python version is compatible with Windows, macOS, and Linux. It is also planned to support Arm Linux in the future.

pip3 install ailia_tflite

Models

Background removal

	Model	Reference	Exported From	Supported Ailia Version
	u2net	U^2-Net: Going Deeper with Nested U-Structure for Salient Object Detection	TensorFlow	1.1.0

Depth estimation

	Model	Reference	Exported From	Supported Ailia Version
	Midas	Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer	Pytorch	1.1.7

Face detection

	Model	Reference	Exported From	Supported Ailia Version
	BlazeFace	PINTO_model_zoo	TensorFlow	1.0.0

Face recognition

	Model	Reference	Exported From	Supported Ailia Version
	Face Mesh	PINTO_model_zoo	TensorFlow	1.0.0
	face_classification	Real-time face detection and emotion/gender classification	TensorFlow	1.1.1

Hand recognition

	Model	Reference	Exported From	Supported Ailia Version
	Blaze Hand	PINTO_model_zoo	TensorFlow	1.0.0

Image classification

	Model	Reference	Exported From	Supported Ailia Version
	MobileNet	MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications	Keras	1.0.0
	MobileNetV2	MobileNetV2: Inverted Residuals and Linear Bottlenecks	Keras	1.0.0
	ResNet50	tf.keras.applications.resnet50.ResNet50	Keras	1.0.0
	EfficientnetLite	efficientnet-lite-keras	Keras	1.0.0
	SqueezeNet	keras_squeezenet2	Keras	1.0.0
	vgg16	VGG16 - Torchvision	Pytorch	1.1.7 for int8, 1.1.9 for float
	googlenet	GOOGLENET	Pytorch	1.1.10

Image segmentation

	Model	Reference	Exported From	Supported Ailia Version
	DeepLabv3+	PINTO_model_zoo	TensorFlow	1.0.0
	HRNet-Semantic-Segmentation	HRNet-Semantic-Segmentation	TensorFlow	1.1.0

Object detection

	Model	Reference	Exported From	Supported Ailia Version
	YOLOv3 tiny	tensorflow-yolov4-tflite	TensorFlow	1.0.0
	YOLOX	YOLOX	Pytorch	1.1.1
	EfficientDetLite	PINTO_model_zoo	TensorFlow	1.1.3

Pose estimation

	Model	Reference	Exported From	Supported Ailia Version
	pose_resnet	Simple Baselines for Human Pose Estimation and Tracking	Pytorch	1.1.7 for int8, 1.1.9 for float

Super resolution

	Model	Reference	Exported From	Supported Ailia Version
	ESPCN	Image Super-Resolution using an Efficient Sub-Pixel CNN	TensorFlow	1.1.0
	srresnet	Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network	Pytorch	1.1.10

Options

You can benchmark with the -b option. You can use the official TensorFlow Lite with the --tflite option.

Launchar

You can use cui launchar.

python3 launchar.py

i.MX8 Support Status

See NXP