Recent Topics

I am currently working with metis M2 card. Trying to run yolov8 model using inference.py file from the sdk. The throughput is very low. I set the AIPU core to 4 and verfied the same in the compile_config.json file. But there is another argument called multicore_mode("multicore_mode": "batch"). what exactly does the multi core mode do? and is there way to set it to other option than ‘batch’?

I need a public library auditorium Pic modern and vintage 

Hi guys. Just a quick note about the deadlines. Due to a sudden convergence of judges being on holiday (which I should have noticed before, but didn’t 😆) it looks like we’re going to struggle to get a judging panel together for the project pitches in that first week of August.So we’ve extended it to the week after, and added a week to the final submissions deadline to accommodate it. 👍Wanted to let you know, and by the way, now you’ve a bit more time you’re welcome to submit more than one pitch, if you have a few projects on the go that you think Wingman might be great for - no reason to only post one idea if you’re cooking up something else that’s equally awesome!

On-device visual ROI tracking - SiamFC on Metis M.2Backstory:It has been on my wish-list since last year to port a visual ROI tracking algorithm on Metis. I have posted about this before.To be clear, visual ROI tracking algorithms are those that do not require an object detector, meaning you should be able to point them at anything and they track the object inside selected ROI (region of interest) in subsequent frames. You can imagine these are very useful for cases when you encounter objects that are not in the trained classes of your typical detection models.Significance:Since Metis is neither a general-purpose processor nor a GPU, accelerating typical OpenCV tracking algorithms on it is not feasible. Fortunately, researchers have developed neural networks for a wide range of tasks, including tracking. Metis is small, incredibly fast, and highly efficient, and its multiple cores can run several models simultaneously. So when it's already part of the system, why not use it to accelera

Project Idea: AI-Powered Multi-Camera Face Recognition & Smart Attendance SystemI want to build a real-time, edge-AI based face recognition and attendance system using Axelera AI Metis and Voyager Wingman.The main goal is to create a system that can process live RTSP streams from multiple CCTV/IP cameras, recognize registered people, track them across frames, and automatically maintain attendance records — while performing AI inference on the Axelera Metis accelerator instead of relying heavily on the host CPU.### What I want to buildThe system will have the following pipeline:RTSP CCTV Camera        ↓Video Processing        ↓Face/Person Detection        ↓Face Recognition        ↓Object/Face Tracking        ↓Identity Matching        ↓Attendance Recording        ↓Web DashboardThe system will initially support one camera and then be designed to scale to multiple RTSP cameras.### Key Features• Real-time face detection and recognition• Multiple registered people• Multiple reference ima

Hi Axelera team and community! 👋I work in retail and F&B commercial operations, so I spend my life around shelves.And there is one question every brand and every retailer wants answered in realtime but nobody can answer affordably:> Is my product there, in the right place, at the right price — right now?Out-of-stock alone is estimated to cost the industry tens of billions a year,and most of it goes undetected because shelf observation is still done withclipboards and checklists. Field audits are slow, subjective and infrequent.Price and promo execution is decided at HQ and forgotten at the shelf.**So here's what I want to build with Voyager Wingman: ShelfPulse.**A compact camera device placed in front of a category runs a computer-visionpipeline on an Axelera system that turns every shelf photo into a structuredaudit in seconds, answering four questions:1. **What's on the shelf?** — every product detected and identified2. **Where is it?** — position checked against the target p

1. Project Overview As global populations age, millions of seniors live independently or spend long hours alone at home while family members are at work. Accidents like sudden falls or nocturnal disorientation pose severe health risks if not detected and responded to immediately.HomeCare-Wingman is a privacy-first, multimodal Edge-AI safety ecosystem designed for senior homecare. Powered by Axelera Metis, the system combines real-time vision pose estimation, local voice command recognition, smart IoT room actuation, wearable sensor fusion, and instant mobile incident management to prevent falls before they occur and deliver immediate emergency alerts when they do. 2. The Problem with Current Solutions Existing cloud-connected home security cameras suffer from processing latency, internet dependency, and severe privacy invasion (seniors strongly object to having raw video feeds of their bedrooms uploaded to third-party cloud servers). Furthermore, single-sensor systems suffer from high

Dear Axelera Support, I am attempting to install Voyager SDK on the radxa rock 5T board with the axelera AI Accelerator M.2. While the SDK suggests using Ubuntu 22.04, I could not find a version of ubuntu for the 5T (https://github.com/DHDAXCW/ubuntu-rockchip-rk3588?tab=readme-ov-file)  I have reviewed previous threads but either they’re not fixed (https://community.axelera.ai/support-central-47/subject-metis-aipu-detected-on-pcie-but-not-responding-stage0-load-failure-on-rk3588-compute-board-voyager-sdk-1-5-3-1280) or there’s an ubuntu version available (https://community.axelera.ai/metis-pcie-7/pcie-connection-doesn-t-work-with-radxa-rock-5b-1101)  I’m not exactly sure if I can use the ubuntu image meant for the Radxa Rock 5B+ on the Rock 5T? radxa@rock-5t:~$ cat /etc/os-releasePRETTY_NAME="Debian GNU/Linux 12 (bookworm)"NAME="Debian GNU/Linux"VERSION_ID="12"VERSION="12 (bookworm)"VERSION_CODENAME=bookwormID=debian radxa@rock-5t:~$ cat /sys/class/metis/version1.4.16 Installation of S

The ProblemMore than 2.2 billion people live with vision impairment. Current assistive tools either provide very limited information (white cane), rely on the cloud (raising latency and privacy issues), or lack robust real-time hazard detection — especially for moving objects, overhead obstacles, sudden drop-offs, and complex indoor/outdoor environments. Users need a system that can detect dangers before they become critical, describe the scene, and do it entirely offline. The IdeaSightCompanion is a fully on-device assistive vision system built with Voyager Wingman on Axelera Metis hardware. Its primary focus is real-time multi-layer hazard detection and prioritization, combined with contextual scene understanding and audio feedback. Core capabilities:- Multi-class hazard detection (moving people/vehicles, static obstacles, overhead obstacles, floor-level hazards, sudden elevation changes)- Distance-aware risk prioritization (closer + faster-moving objects get highest priority)- Seman

I run digital transformation workshops for Italian SMEs, which means I spend a lot of time in meeting rooms covered in sticky notes. Process mapping sessions, mostly. And after years of trying every digital tool out there, I've accepted the truth: pen and paper wins. People think better on a whiteboard. They sketch flowcharts, build effort/impact matrices with sticky notes, draw arrows everywhere. It works.What doesn't work is everything that comes after. Someone (usually me) photographs the wall, and then spends an evening squinting at pictures, retyping notes, trying to reconstruct a matrix that made perfect sense in the room. Regular OCR doesn't help much either, because it gives you the words but loses the structure. A matrix without rows and columns is just a word list. So in practice, half of that knowledge dies when the board gets erased.So here's what I want to build: a small box that sits in the meeting room and captures the knowledge while it's being created.A camera points a

I see to many AI right now. What the golas of Axelera ai?

我是一名即將步入法律系大三上的法律系大二生，但不知道要如何讀好書

Project ideaI want to build Avertaq SiteGuard, an edge AI safety and compliance assistant for construction sites.The system will process a live camera feed locally and detect:- workers entering a predefined restricted or hazardous area,- workers not wearing a safety helmet,- workers not wearing a high-visibility vest.When a compliance event is detected, the system will save a timestamped event and a relevant image, rather than continuously uploading the full video stream. It will then generate a simple daily safety and compliance report.Where it will be usedThe project is designed for construction sites, warehouses and other industrial areas where safety monitoring is currently performed manually or through continuous CCTV observation.The initial prototype will use one camera and one predefined monitoring zone.Why it is usefulConstruction companies need a practical way to improve safety monitoring, document incidents and produce evidence for internal compliance reporting.Running the in

The Problem, Personally As a drone builder and hardware practitioner, I have spent a lot of time developing and testing different aerial platforms, including FPV systems, while working with low-latency video transmission and real-world flight environments. In my practical experiments with airborne anomaly detection using classical computer vision (standard contour, edge, and motion detection), I ran straight into a frustrating limitation: false alarms caused by dynamic sky environments.When a camera is pointed at the sky from a moving platform, classical edge detectors treat moving cloud boundaries, changing light textures, and sun flares as airborne objects. The system constantly generates false positives. For real-world drone operations, aerial inspection, or pilot awareness, simple pixel-contrast heuristics are simply not reliable enough.To evaluate real physical objects against a dynamic sky background, simple pixel boundaries aren't enough — you need local Edge AI inference capabl

FloorSight — a perception brain for the open-source robot vacuum [The Prompt submission]The problem, personally.My Roomba is old. Years of service have worn it down, and it still fails the same two ways it always has. First: it eats things. Charging cables, phone chargers, shoelaces — anything below the height its sensors can see just disappears under it until something jams. Second, and worse: there is one room in my house it must never disturb — our prayer room — and it barges in anyway, bumps into things, knocks items around, and gets itself stuck. App-drawn no-go rectangles don't cut it, because the robot doesn't actually see the room; it just trusts a map that drifts.This isn't just my Roomba. The robot vacuum market ships over 12 million units a year, is worth more than $12B, and is still accelerating — IDC counted global shipments up 18.7% in the first three quarters of 2025 alone. Yet the dominant sensing stack — bump sensors plus a 2D LiDAR turret — is structurally blind below

Hello Axelera AI Community!I'd like to participate in The Prompt Challenge with a project called AquaRov AI.Marine aquaculture is one of the fastest-growing food industries, but underwater monitoring is still largely performed manually. Fish farmers often face problems such as damaged nets, poor water quality, fish disease, biofouling, and fish mortality before they are aware of the issue. These problems reduce productivity and increase operational costs.My goal is to build AquaRov AI, an Edge AI-powered underwater monitoring system using an ROV (Remotely Operated Vehicle), underwater cameras, and environmental sensors running on Axelera Metis.The system will continuously inspect fish cages and analyze underwater conditions in real time. Instead of only recording video, AquaRov AI will use AI to understand what is happening underwater.The AI system will be designed to:Detect damaged or torn fish nets.Detect dead or unhealthy fish.Monitor fish behavior.Detect excessive biofouling on fis

The ProblemAutonomous vehicles and robots need to understand the world in 3D. The industry standard is BEVFusion, fusing multiple cameras and LiDAR into a unified bird's-eye-view representation for 3D object detection. Every production implementation runs on NVIDIA hardware. The cost, power draw, and vendor lock-in are significant barriers for robotics at scale. Example NuScenes Dataset with 6 cameras, 1 Lidar to get 3D detections and Segmentation for ADAS pipelineWhat I Want to BuildA fully working BEVFusion pipeline on a single Axelera Metis PCIe card. Six camera backbones, LiDAR processing, BEV feature fusion, and a 3D detection head, running in real time without any NVIDIA hardware.The model is FastBEV https://github.com/ymlab/advanced-fastbev), a CNN-only BEV fusion architecture with no transformer attention, designed specifically for efficient hardware deployment. No attention means no DPU fusion issues. Pure Conv-BN-ReLU throughout.The deployment plan:Six camera backbones run on

Developed for czela.net (a non-profit community ISP in the Czech Republic with ~400 active infrastructure devices and 2,000 member connections) this project bridges real-world physical network reality with NetBox SSoT.            Standard telemetry tools (SNMP, LLDP, FDB, OSPF) often fail due to incomplete data and unmanaged hardware cascades. Our hybrid agent combines standard sync code with a local quantized LLM. It generates practical          recommendations for volunteer network admins to fill documentation gaps, assists in local troubleshooting during outages, and preserves complete data privacy without cloud dependencies.Rather than replacing existing monitoring software, our project introduces a **Hybrid Auto-Discovery & Reliability Agent**. The core logic is **hardware-agnostic** and designed to run local LLMs (currently prototyped with Ollama/mistral-nemo). We are interested in evaluating **Axelera Metis** as a potential low-power edge accelerator for our network distribu

This project goes one level deeper than the typical AI pipeline flow, where several pre-compiled models are assembled into an application.The current Autoware Perception stack (Vision Pilot :  https://github.com/autowarefoundation/vision_pilot) was designed for NVIDIA, and does not compile for Axelera Metis.This is today’s reality, a situation that engineers often face when moving an algorithm from NVIDIA to more efficient embedded hardware.This project will explore the use of Wingman to identify the problematic unsupported layers, modify the model architecture, and re-train the models for Axelera Metis.If successful, this will be ground-breaking … an Edge AI NPU that can port a model developed for NVIDIA to itself !If not successful, this will be a good use case for further exploration by the Axelera innovation team.Cheers !Mario Bergeron (A.K.A. AlbertaBeef)

Hello, I am a new user. I joined the platform but I cannot find any AI tasks. Can someone guide me on how to access tasks and start working? Thank you.