R&D 연구(LVM)

Research

Vision AI

Vision AI
is an artificial intelligence technology that
helps computers deeply understand and
interpret the visual world. It can analyze
digital images and videos to accurately
identify and classify objects.

Progress in Vision AI is fueled by
improvements in hardware, massive
amounts of data, algorithms, deep learning,
and other technologies that are spurring
innovation in various sectors like
manufacturing, healthcare, automotive,
security, and beyond.

Advancement

Current state

Industry
transformation

History of XIIlab
and Vision AI

2010~2012

XIIlab Establishment

AlexNet wins image recognition
competition which spotlights Vision AI

2012~2013

Acquire big data processing technology

Convolutional neural networks and others
improved model accuracy

2013~2017

Start developing data-driven AI technology

AlphaGo’s victory boosts interest in AI

2017~2019

Gain the ability to commercialize AI video

Expanding the Vision AI market

2019~2020

Join NVIDIA Partner Network

Expanded use of technologies
such as autonomous driving

2021~Present

Listed on KOSDAQ and incorporated
in the US

Practical applications in multiple fields

The evolution of the
Large Vision Model

Following advancements from Texture
to Vision Transformer, SAM, and GPT,
the next stage in the Large Vision Model
is Computer Vision.

Fixed environment recognition

- Texture
- Image Pattern

Learn data patterns

- Deep Learning
- Convolutional Neural Networks (CNNs)
- Layer
- Image Classification
- Object detection
- Segmentation

Attention is All You Need

- Transformer Architecture
- Reduced training time
- Effective long-sequence handling

GPT, Transformer Base

- Google, Vision Transformer, VIT - image
sequence
- CLIP
- Effective long-sequence handling

Multi Modal : LVLM

OpenAI (DALL-E, GPT-3, GPT-4)

META, Segment Anything, SAM

Large
Vision-Language
Model, LVLMs

Multi-Modal Learning

Combines both pictures and words to
obtain more valuable and practical
information.

Detailed and accurate situational awareness

Achieve improved situational awareness
by combining text descriptions and visual
content for better accuracy and detail.

Developing creative applications

Develop new forms of creative
applications, such as art generation,
descriptive captions, conversational AI,
and more.

GPT-4 (OpenAI)

Language models for
creativity, visual input
processing, and long
contextual understanding

DALL-E 3 (OpenAI)

Models that can convert
text to images

SORA (OpenAI)

Models that can convert
text to video

Stable Diffusion
(stability.ai)

Image generation model