//
Search
🇹🇩

June- CVPR 2023

Hello everyone! We’re already halfway through the year as we’ve been busy during the first half of the year! There is a conference that comes around this time every year and it is CVPR! Following last year, LG AI Research has again become a sponsor at this year’s CVPR conference along with many other companies. This year’s CVPR was held in Vancouver, Canada. Compared to the hot and humid weather in Korea, June in Vancouver was surprisingly chilly (while last year’s CVPR in New Orleans was so hot and humid!) Let’s leave the chilly weather behind and dive into the heated scene at the conference!
At CVPR 2023, LG AI Research hosted the New Frontiers for Zero-shop Image Captioning Evaluation (NICE) Workshop in collaboration with the Seoul National University and Shutterstock! The objective of the NICE Workshop is to develop a powerful image captioning model that can advance the latest technology in accuracy and fairness. The challenge before the workshop focused on testing the true limitation of zero-shot image captioning, with 142 global research teams participating.
Shutterstock, a partner of LG AI Research, provided the newest open-source datasets. The provided datasets were composed of about 26,000 high-quality images along with preselected metadata that involves more than 20 general categories with a wide range of concepts! Many participants from various universities and companies around the world joined our workshop.
Honglak Lee, CSAI at LG AI Research and also a professor at the University of Michigan, and many others from global IT companies and universities including Google Research, Microsoft Research, and Allen Institute for AI, participated as the invited speakers.
After the NICE Workshop, the LG AI Day dinner event was held on June 19, the day before the main conference.
LG AI Day is a networking dinner event where we extend an invitation to all the outstanding researchers from universities around the world to learn about the computer vision research of the LG Group and share the latest research trends.
Unlike the previous year, this year’s LG AI Day was divided into two parts. The event officially kicked off with Hailey Jeong, Lead, Talent Relations Team as the host, followed by an opening speech by Honglak Lee, introducing the AI research of LG affiliates.
Then the dinner and networking began and the researchers at each table were engaging in communication regarding their research, etc with each other. This concluded the first part of the event!
Part 2 involved a mix of individuals working on similar research topics and formed a new group for the networking session. We saw more active networking than in part 1, probably due to being in the same research field.
In addition, there were many fun events at LG AI Day like quizzes and a raffle! Many products from the LG affiliates were prepared as prizes (such as the newest StanbyMe GO! ) We received enthusiastic responses from the participants. With the raffle as the final session, CVPR 2023 LG AI Day also came to a successful end. We once again give our sincere thanks to everyone who attended.
With the workshop and LG AI Day behind us, wouldn’t you say the main event of the conference is the company booth?
From June 20–22, the main conference went on for three days. Five LG affiliates participated in this year’s CVPR: LG AI Research, LG Electronics, LG Innotek, LG Energy Solution, and LG Uplus.
Unlike last year’s CVPR, we had open seminars on various topics from each subsidiary at the booth which included Robust Anomaly Detection, Industrial AI Technologies, and Continual Learning. (Thank you to the presenters and everyone who attended to listen! )
There was also a fun demo along with the seminars! We showcased a demo of the photocards and tattoo printing using the images generated with EXAONE, LG AI Research’s large-scale AI model based on text-to-image technique. For tattoo printing, a tattoo design created with EXAONE was transferred to IMPRINTU, the mini tattoo printer developed by LG Household & Health Care, to be printed on the arm or hand of those visiting our booth! (The base powder, ink, and fixer that come into contact with the body are all composed of harmless ingredients)
Finally, a demo was also prepared by the Vision Lab - Deep Document Understanding squad! Material informatics, a new material development using AI, relies heavily on training data, and thus, securing the chemical substance databases through professional technical literature is vital. Researchers have been reading technical literature, drawing molecular structures, or directly inputting chemical properties into the system that posed challenges of slow processes and high costs. Thus, a demo was provided at the booth of technology that distinguishes images, tables, charts, and layouts from technical documents in PDF format, such as papers and patents, which are then analyzed and converted to text data that can be processed in AI!
This technology is based on technical elements of multimodal learning that understand both image and text information, OCR technology that detects and converts images to text, and graph neural network (GNN) that clarifies the connection and relationship between elements! Furthermore, training with training data built by LG AI research enables highly accurate conversion of information such as structural formulas and charts that may not produce satisfactory performance with just existing open datasets. You can tell that CVPR 2023 has successfully ended!
We were able to have an even more successful conference thanks to everyone who visited our LG booth Since the establishment of LG AI Research, we have participated in many graduate school events and research institutes and met with many people in the past two and a half years. It was a delightful and memorable week meeting with people overseas rather than in Korea (It was great to see all the familiar faces and new research members at the conference!!)
Following the previous CVPR, it was a good opportunity for the employees and executives to experience the incredible research achievements on-site. We plan to actively engage and strive to solidify our position as a leading research institute in the global AI technology scene and actively engage with many talented individuals at the AI conferences this year and other upcoming conferences! We hope everyone had a safe flight back and we’ll see you again next time!