XGRIDS’ Post

Great study by Manuel. Next step is to optimize the 3DGS processing times, GPU requirements and enhance the compression for smaller data sizes. Great looking data!

🚀 Unlock the full potential of your AI applications with the AIP-SQ67. Introducing the AIP-SQ67, an expandable AI inference platform crafted for superior graphics and edge AI capabilities. Powered by 12/13th Gen Intel Core™ processors and featuring an extension AI accelerator MXM module slot, this platform is perfect for AI developers and system integrators seeking top-notch performance for their AI tasks. 🔍 Precision AI Vision at the Edge: Ideal for seamless integration into vision-related AI applications such as object inspection, defect detection, and video analysis, enhancing efficiency and ensuring safety in production workflows. 💡 Expand Possibilities: With extensive I/O extensions, including DisplayPort, LAN, and USB ports, and support for NVIDIA® GPU or Hailo-8 MXM modules, the AIP-SQ67 meets the demands of heavy AI workloads and fast image data analytics. 🌐 Effortless Remote Management: Equipped with Aetina's EdgeEye platform, it offers robust remote management capabilities, ensuring swift and convenient maintenance of all connected edge devices. Experience faster processing, quicker insights, and enhanced decision-making right at the edge: https://lnkd.in/dQr-JuSP

Apple released 4M-21 last week – an any-to-any vision-language model 🔥 🔥 🔥 (the video example shows the generation of an RGB image conditioned on the caption and depth map) M4-21 is a 3B parameter model released by EPFL and Apple. It is multitask and multimodal, trained on 21 diverse modalities across 1T tokens. It can predict any modality from any other: image captioning, depth estimation, object detection, instance segmentation, pose estimation, image generation, and more! We can mix modalities, such as generating images based on descriptions and bounding boxes, assigning different priorities to steer the generation, and chaining multiple generation steps. 🔥 Check the project page for details. Key takeaways and links to the paper, code, and HF Space are in the comments below. 👇🏻 ⮑ 🔗 project page: https://4m.epfl.ch #computervision #transformers #multimodalai #opensource

Vision is all you need

Our soon-to-be-launched Video Management Server for Polytron.AI cameras. POLYTRON ONE is our UNIFIED Video Management Server connecting Polytron and non-Polytron cameras, with an integrated LIVE 3D AUTONOMOUS virtual patrol and other security surveillance feature. It also features our SEMANTIC 3D engine, a zero-shot generalisation segmentation AI Model that understand any and every "object" in our LIVE 3D scene without need for more training on new objects. We are deploying this with one of the world's LARGEST security agencies. Watch this space with updates. 2500 POLYTRON cameras and 20 POLYTRON ONE servers. ## 3 KILLER Features ## - Integrated **LIVE 3D Viewer.** Click on any video panel that is connected to Polytron Camera, you’d zoom in and navigating in 3D space in REAL-TIME. - Connect up to **250 or more 4K 360 video streams** (Polytron Cameras) using just an INTEL NUC or NVIDIA Orin EDGE Server Box. Don’t need server-grade INTEL GPU nor NVIDIA high end GPU server. - Experience **BLAZING-FAST VIDEO SEARCH** through "**3D Semantic Scene Understanding**". We use natural language, either voice or text, to search the recorded video archives. Our AI can sift through gigabytes of video content to locate the exact frames you're looking for. The 3D Semantic Scene Understanding Large Language Model (LLM) possesses the capability to recognise and understand millions of distinct objects. And we are training more objects as we speak. Right now we can recognise 1.5m objects. Demo: https://lnkd.in/gNc5tNs6

OPT Industrial Camera " One Stop Solution" Machine Vision. The camera has a resolution range of 300,000 to 150 million, a complete data interface, a compact structure, comprehensive functions, and excellent imaging, meeting various application scenarios. Why use a PC-based machine vision system when Smart cameras seem to dominate the vision world? While each technology has its place in the market PC-based systems may offer more than you think. In the 1990s the progression of smaller chips and better circuit board technology gave us a cost point that allowed the birth of the Smart camera. Now manufacturers could put a processor, imager and I/O in one small package. All of a sudden road blocks such as size and cost were not factors on the factory floor. While the Smart camera revolution did greatly expand the vision market, it did not replace all PC-based systems. PC-based vision systems generally consist of a PC with a Windows operating system. A frame grabber or Ethernet card in the PC interfaces to cameras to import the images. A software package on the PC analyzes the images, makes decisions and sends data such as pass/fail information to another system. Sound like 1987? That’s because it is. The hardware behind PC-based vision systems has changed very little over the last 20 years. The difference is that changes in computing power now allows for more powerful software. Following are some examples of what the computing power of PC-based vision can do.

💥💥💥 REDUCIO! Generating 1024×1024 Video within 16 Seconds using Extremely Compressed Motion Latents Rui Tian, Qi Dai, Jianmin Bao, Kai Qiu, Yifan Yang, Chong Luo, Zuxuan Wu, Yu-Gang Jiang Abstract Commercial video generation models have exhibited realistic, high-fidelity results but are still restricted to limited access. One crucial obstacle for large-scale applications is the expensive training and inference cost. In this paper, we argue that videos contain much more redundant information than images, thus can be encoded by very few motion latents based on a content image. Towards this goal, we design an image-conditioned VAE to encode a video to an extremely compressed motion latent space. This magic Reducio charm enables 64x reduction of latents compared to a common 2D VAE, without sacrificing the quality. Training diffusion models on such a compact representation easily allows for generating 1K resolution videos. We then adopt a two-stage video generation paradigm, which performs text-to-image and text-image-to-video sequentially. Extensive experiments show that our Reducio-DiT achieves strong performance in evaluation, though trained with limited GPU resources. More importantly, our method significantly boost the efficiency of video LDMs both in training and inference. We train Reducio-DiT in around 3.2K training hours in total and generate a 16-frame 1024*1024 video clip within 15.5 seconds on a single A100 GPU. 👉 https://lnkd.in/dV49Ux-a #machinelearning

Industrial enterprises — the automotive, manufacturing and electronics organizations of the world — are in pursuit of digitalization, but are held back by mounting data silos and collaboration challenges. In response, a growing number of organizations across the world’s largest industries are unlocking interoperability across a complex ecosystem and throughout the manufacturing pipeline with the open-source Universal Scene Description (USD) standard, or OpenUSD, for an expanding variety of use cases. “USD is the closest thing to a universal 3D standard there is today, but it’s currently only a de facto standard, not an internationally recognized standard like PDF or HTML,” says Guy Martin, director of open source and standards at Nvidia. “Industries like manufacturing, construction, architecture and engineering want and need to have open standards that they can count on.” https://lnkd.in/grXUd3de #techarticle #monday

Designers and engineers will be the next big beneficiaries of AI development. AI-optimized technologies, such as NVIDIA RTX can handle a large amount of detailed visualization and modelling required for these industries. While Lenovo Workstations are becoming increasingly sophisticated when it comes to supporting next-gen tools like 3D modelling and AI inferencing. I’m interested to see what comes next – particularly the impact of tools like Stable Diffusion for rapid image generation and intelligent visualization to support design and engineering tasks. Learn more about AI development and Lenovo Workstations: https://rebrand.ly/lwx8x18

It is my responsibility as a writer to produce engaging, informative, and persuasive content that appeals to the audience I am targeting. Conveying information effectively and aligning the content with the company's aims and voice are the same objectives when writing blog posts, social media updates, website text, or marketing materials. Research and creativity must be used to create information that is accurate, timely, and interesting. It is essential to retain coherence and clarity when adapting writing style for different platforms and readers in order to boost reader engagement and assist customers or employers in achieving their goals.