|
Greetings everyone! 👋🏻
In this issue of our newsletter, we will focus on the latest product release from Nota AI, PyNetsPresso, launched in June. We'll also share updates on our participation in other global events, including EMP (Entrepreneur Market - Fit Program) 2023 and Silicon Valley June Summit 2023.
Let's dive in together. 🌊
|
|
NetsPresso®, the hardware-aware AI model optimization platform from Nota AI, is released as a library using Python code! 🙌🏻
PyNetsPresso accelerates artificial neural networks and provides usage code-related repositories that help you develop optimized AI models more efficiently! 💥
It provides the necessary code repositories related to the task on GitHub. By utilizing this code, users can develop their models into optimized AI models! 👏🏻
Streamline your AI model optimization workflow with PyNetsPresso! 🚀
What PyNetsPresso offers
1️⃣ Model Zoo
- Image Classification
- Object Detection
- Semantic Segmentation
2️⃣ Compress
- Structured Pruning
- Filter Decomposition
3️⃣ Convert (to be released)
- Supported devices
- Renesas RZ/V2L, RZ/V2M, RZ/V2MA
- Sony IMX500
- NVIDIA Jetson
- Raspberry Pi
- Intel Xeon
|
|
Using the newly unveiled PyNetsPresso, the AI model achieves the best possible performance while maintaining high FPS. 👏🏻
For instance, PyNetsPresso accelerates the inference speed of YOLOv5-L by 2.6× without sacrificing mAP performance. 🤖
For more details, read the article to learn how PyNetsPresso can revolutionize your AI optimization workflow.
|
|
✅
Approved for ICCV 2023 Demo Track
|
|
|
We are thrilled to share that two of Nota AI's demonstration tracks have been approved for ICCV 2023 (International Conference on Computer Vision). The ICCV is a prominent global conference focusing on computer vision, which makes the inclusion of two demos from Nota AI, especially noteworthy. Let's get into the specifics! 👀
|
|
|
Accelerating Speech-Driven Talking Face Generation with 28× Compressed Wav2Lip
|
|
The demonstration presents a streamlined model that creates a speaking face from speech. By eliminating residual blocks and minimizing channel width within the standard Wav2Lip, a more compact generator for fabricating speaking faces was developed.
By employing selective quantization, it accomplishes substantial speed enhancements up to 19 times faster on edge GPUs while maintaining the quality of the produced results.
|
|
Architecturally Compressed Stable Diffusion for Efficient Text-to-Image Generation
|
|
The demonstration presents a streamlined Stable Diffusion Model (SDM) for versatile text-to-image generation. Nota AI's BK-SDM-Small has received considerable recognition for its ability to decrease parameters and latency by 36%. Even with scarce training resources, Nota AI's model distinguishes itself by utilizing the acquired knowledge from the primary SDM to generate high-quality replicas.
|
|
📃
Conferences/Presentations
|
|
|
|
On-device Intelligence Workshop at MLSys 2023
|
|
Nota AI participated in the On-device Intelligence Workshop at MLSys 2023, held at the Miami Beach Convention Center from June 4th to 8th.
Our researchers Bo-Kyeong Kim and Junkyeong Choi, presented their latest research on "A Unified Compression Framework for Efficient Speech-Driven Talking-Face Generation" and "Arithmetic Intensity Balancing Convolution for Hardware-aware Efficient Block Design." Their presentations were truly remarkable and left a lasting impression! 🔥
|
|
|
AI City Challenge Workshop at CVPR 2023
|
|
Nota AI participated in AI City Challenge Workshop at CVPR 2023 held at the Vancouver Convention Center from June 18th to 22nd.
Wooksu Shin, Research Engineer at Nota AI, and Jeongho Kim, Engineer at Nota AI, presented our paper, "Addressing the Occlusion Problem in Multi-Camera People Tracking with Human Pose Estimation," as a poster on-site. 🗣️
We also had the privilege of hosting a live Q&A session, where we engaged with participants, shared insights, and answered thought-provoking questions related to our research. It was truly an engaging experience! 💡
|
|
|
ES-FoMo Workshop at ICML 2023
|
|
We are excited to inform you that Nota AI's paper has been approved for presentation at the "Efficient Systems for Fundamental Models" workshop at ICML 2023! ICML is globally recognized as a premier conference in the machine learning sphere, and we are pleased to bring you this great news.
The approved paper, "BK-SDM: Architecturally Compressed Stable Diffusion for Efficient Text-to-Image Generation," delves into optimizing and compressing the crucial Stable Diffusion model, a key element in advancing Generative AI.
If you are interested in observing the functionality of this technology in action, please take a moment to explore the Hugging Face demo space provided below. 🤗
|
|
|
EMP (Entrepreneur Market - Fit Program) 2023
|
|
Nota AI participated in the EMP 2023, which took place in Silicon Valley, USA, from April 24th to May 12th, orchestrated by KIC. EMP is a program specifically created to provide support for startups looking to break into the North American market. It includes lectures, opportunities for networking, mentorship, and FalconX sessions. Notably, Nota AI's co-founder Tae-ho Kim led the presentation, “Hardware-aware AI model optimization specialists,” garnering significant interest.
|
|
|
Silicon Valley June Summit 2023
|
|
During the Silicon Valley June Summit 2023, which took place in Silicon Valley, Steven Kim, Co-CEO of Nota AI, delivered a keynote speech. At this event, Nota AI proudly unveiled two flagship products, NetsPresso® and Nota DMS.
The Silicon Valley June Summit 2023 event is recognized for showcasing over 145 startups across diverse sectors such as AI, new materials, mobility, and cryptocurrencies. It also hosts a variety of speakers, thereby strengthening our standing in the startup community.
|
|
|
|
