We support R&D and product development by utilizing our high technology
Advanced AI Research Support
Utilizing Araya's high technological capabilities such as deep learning, autonomous AI, and neuroscience, we support various research and product development related to AI.
|Communications Industry||Human detection and semantic domain segmentation model from Full HD images captured from a drone flying at 30-40m altitude, compressed and implemented on Jetson TX2, FPGA, etc. with a common base net.|
|Automotive Industry||Implemented a model on Raspberry Pi 4 + Edge TPU that detects a person up to 20 m away and projects it on the world coordinate system to estimate the position.|
|Research and development to improve the accuracy of the object detection model after compression by distillation.|
|Developed a model that divides the drive recorder video into more than 15 classes of semantic areas such as automobiles, pedestrians, and signals.|
|Development of an algorithm that detects obstacles with innumerable patterns from in-vehicle camera images without a correct label.|
|Driver behavior estimation algorithm in the car. Classify behaviors such as eating and drinking and talking by combining with posture estimation.|
|Plant Industry||Labor saving of indoor inspection by autonomous drone.|
|Game Industry||Automation of replay / debugging with reinforcement learning|
Edge AI Consulting
We will propose the model selection, compression method and appropriate hardware combination according to your accuracy and speed requirements.
We also undertake the actual model training, compression, and hardware implementation work.
1. Total Solution Service
Also, if necessary, it is possible to enter from the requirement definition and carry out.
We will take care of everything.
2. Individual Consulting Service
Requirement Definition Support Service
We will assist you in defining requirements from multiple perspectives for your purposes.
Edge Implementation and Optimization Services
Convert the deep learning model to a model suitable for the edge environment you want to mount, and evaluate the speed using a real device.
Model Miniaturization Service
After looking at the model, input / output data, learning code, etc., we will propose a method for making the model smaller. We can also implement it upon request.