AI research is being carried out across various fields such as Image classification, Object detection, and Pose estimation. AI models are becoming more advanced day by day, and synthetic data is also gaining popularity as the need for high-quality training data has emerged. Therefore, it will be possible to improve the accuracy of AI models that have already been developed and provide optimal synthetic data for AI models that have limitations in collecting real data, thereby providing more accurate and faster results to AI model developers.

Fine-Grained Image Classification

Robust AI models can be developed by classifying them in more detail than conventional coarse-grained image classification and utilizing domain-randomized data. The fine-grained image classification AI model is used in various fields such as automobiles, ships, and plant variety.

Object Detection

By localizing and classifying the specific object within the image through the AI which trained by auto labeled data, it rapidly and easily helps the vehicle model identification system and the object tracking system.

Pose Estimation

Enhances the accuracy of AI models using 99.9% precision 2D & 3D key point data collected from the motion capture system. This is an area of AI in where you extract 2D, and 3D key point data of persons in an image and use it to estimate their posture. It is commonly used in sports analysis, healthcare, abnormal behavior analysis, etc. where precise recognition of movement is required.

Synthetic Data validation

For validating an effectiveness of synthetic data, we tested a car data set classified by 50 models with an image classification AI. At first, we trained an AI with gathered images of cars from google only. (Each of them has 50 pictures.) Then, it results 76.02% of accurac y. Secondly, we trained an AI with bunches of real images same as the above and synthetic data. It results 90.94% of accuracy. it shows that synthetic data can be used for building an advanced AI model and it is useful to enlarge a data set with limited data.

Train data

Test data

Result