Please enable JS

Life with Deep Learning

We are bringing Deep Learning technology to every corner of the world,
creating a richer and more creative society.

What We Do ?

License Business

We provide a framework to make deep learning compact and simple. Using our unique libraries, it becomes possible to compress large and complex models and run them even on low-powered edge devices such as FPGA, CPU, etc.

Service Delivery

We provide Deep Learning SaaS, enabled by our platform "DeLTA-Lite". This makes deep learning easily accessible for everybody.

Solutions Business

We offer deep learning solutions for various industries and applications based on LeapMind's unique technology and expertise.

Use Cases

Application for autonomous driving

Detecting the position of other vehicles and identifying traffic lights, signs, etc., to achieve autonomous driving. Since computation is done on edge devices, it becomes possible to integrate them without requiring large computational units.

Structural anomaly detection in factories and buildings

WBased on images taken by robots and drones, we can automatically detect structural defects and anomalies in factories, buildings, etc. where manual inspection is difficult.

Edge intelligence and Internet of Things

We enable the use of deep learning on IoT devices such as smart home appliances. Making use of highly efficient edge processing, we make life more convenient with applications such as recipe recommendation and automatic room light adjustment.

Software Technology

Deep Learning compression and acceleration

We develop highly efficient Deep Learning architectures that work with limited memory and at high speed, customizable to meet specific hardware requirements.

500x model compression while maintaining performance

Compressed models can be implemented on to embedded devices. This reduces the cost of Deep Learning.

Performance

(test: metrics/accuracy)

Size

Power efficiency improvement

Compressed models can be further optimized through FPGA acceleration, resulting in faster computation with less power consumption.

Image Classification

Image Segmentation

Hardware Technology

Hardware Design

Our embedded Deep Learning solutions make use of an unique hardware IP that can be implemented on small size FPGA-chips.

10x speed up with Deep Learning on FPGA

By constructing a circuit dedicated to Deep Learning on FPGA, we can provide 10x speed up as compared to CPUs.

Processing time per frame

Image Classification

Network: classification.lmnet

Dataset: cifar10

Image Segmentation

Network: segmentation.segnet

Dataset: segmentation