seeed studioCoral USB Accelerator Accelerator coprocessor for Raspberry Pi and Other Embedded Single Board Computers

Dropped my inference times to about 12 ms. My home server doesn't have a beefy video card, so it was very slow using just the CPU. This was a cheap and easy way to speed up my image recognition without access to a PCI slot or spending a lot more.

I bought a Raspberry Pi 4 and 5 to run an NVR and quickly found that they are nowhere near powerful enough. Fortunately, this product exists. It does all of the heavy lifting and makes this entire configuration possible. Had I known beforehand, I probably would have gone a different route, but having already acquired the hardware, I'm glad such a solution exists.

Arrive quite quickly. Plugged it in using supplied cable to USB port on back of a Topton fanless 4 port mini PC (n5105) and passed through USB port from Proxmox to Frigate docker container (running in Docker LXC). Set up camera (Reolink RL-810A) and checked Inference Speed of Coral TPU and it was not good at 29.3 ms. Checked USB port specifications of Topton mini-PC and remembered that the back USB ports are all just USB gen2. Both USB A type ports on front of the Topton are gen3 but both were already in use so plugged Coral cable into a USB-A to USB-C adapter and then plugged adapter into front USB-C port on Topton mini PC. Re-configured USB pass through from Proxmox to container and restarted. Inference Speed dropped to 9.2 ms - should have read the section on using fastest USB port! Added 3 more cameras (one more Reolink RL-810A, one Eufy 2K indoor and one Eufy 2k Pan and Tilt indoor). Inference Speed looks stable at about 8.9 ms (not as good as some reports in frigate forums but still good given Coral is passed through from Proxmox to Docker LXC then Frigate docker container inside that).Coral TPU has been working well for past week and false positives cut down markedly from Eufy cameras (mostly cat detection) compared to native camera app's 'pet' detection while notifications from Frigate NVR via Home Assistant to phone are a few seconds faster than the notifications from the native camera apps but not much in it. Person detection from Reolink itself seems on a par with frigate/coral however but Frigate + Coral is work well with Person, cat, dog, bicycle and motorbike objects while Reolink natively only has Person, Vehicle and Pet.Overall pleased with Coral unit (even though seems expensive) and very pleased with speed of dispatch as well as delivery time from Seeed Studio Store.

Both devices I received, the original and it's replacement, were completely defective. Neither one would ever power on despite trying no less than 10 different USB cables, two different powered USB hubs, two laptops, two desktops, one workstation, and one server, collectively running Windows 10, Windows 11, MacOS, and various versions of Linux. Safe to say that the devices were DOA. Needless to say, there won't be a third time. Please learn from my lesson and don't waste your money or your time.

Configures easily for Frigate for object detection, even if using more than one. Can get a little hot! Ensure it's not competing with other high bandwidth devices on the same USB bus (some devices might be internal)

This little device doe ai detection then even high end server cpus And barely uses any power. Latency for detection in Frigate is very low.

As the headline states. I am currently using this with Frigate running on home assistant os running in a VM on Truenas. I had to pass through my whole USB controller as a PCI device. This is ok for me since the only other thing plugged in is a Sonoff USB Zigbee gateway. Product works perfectly and without hassle once adding in the correct detector config to frigate.yml.Product is working great! I only have one camera for the time being. Went from 110ms response time to 11ms response time. I'm excited to now expand the cameras around my home.I just wanted this information out there in case somebody else feels like it's possibly not still supported based on other reviews here. A little bit of searching on the internet will show how to pass through to VM. I am by no means an export and very much new to the VM world.

I am thoroughly impressed with the Coral USB Accelerator—it's truly a game-changer for AI camera systems! As a developer working extensively with AI applications for surveillance and monitoring, this device has exceeded all my expectations.Firstly, the setup was incredibly straightforward. I simply plugged the Coral USB Accelerator into my existing camera system and connected it to my development environment. Within minutes, I was up and running, ready to integrate powerful AI capabilities into my cameras.The performance is exceptional. The Coral USB Accelerator significantly boosts the processing power of my cameras, allowing for real-time object detection, classification, and tracking. Even with multiple cameras running simultaneously, the Coral USB Accelerator handles the workload effortlessly, ensuring smooth and accurate AI inference.What I appreciate most is the versatility of this device. It supports various AI models, allowing me to choose the best one for my specific application. Whether I'm detecting intruders, monitoring traffic patterns, or identifying wildlife, the Coral USB Accelerator adapts seamlessly to different use cases.The compact size of the Coral USB Accelerator is also a big plus. It's portable and doesn't take up much space, making it ideal for both indoor and outdoor installations. The low power consumption is another bonus, ensuring that my camera systems remain energy-efficient.Overall, the Coral USB Accelerator has revolutionized the way I approach AI camera systems. It has enhanced the intelligence and responsiveness of my cameras, enabling more sophisticated and effective surveillance solutions. If you're looking to integrate AI into your camera network, I highly recommend the Coral USB Accelerator—it's a must-have for AI developers and security professionals alike.