megaAI 4K AI camera board reminds me of Kendryte K210 based boards such as Maixduino used for computer vision for tasks such as object tracking or face recognition, but instead of just handling QVGA at around 15 to 18 fps, megaAI can supports inference at 4K resolution up to 30 fps.
The tiny board can achieve this feat by leveraging the 4 TOPS of AI processing power delivered by Intel Movidius Myriad X VPU (Vision Processing Unit) while consuming only around 2.5 Watts.
megaAI board specifications:
- AI Accelerator – Intel Movidius Myriad X VPU delivering up to 4 TOPS (Trillion Operations Per Second)
- Camera – 12 MP 4056×3040 Auto-Focus (8cm-∞) camera with 81 degrees horizontal Field of View (HFOV)
- RAW Output – 4K @ 60FPS or 12MP @ 30FPS (with USB3.1 Gen2 Host)
- Hardware Image Encoding – 12 MP JPEG encoding
- Hardware Video Encoding: 4K H.265 encoding at 30 FPS
- Host Interface – USB 3.1 Type-C port (5 Gbps)
- Power Supply
- USB-C port
- Power Draw – 2.5 W max
- Dimensions – 45 x 30 mm
- Weight – ~57 grams
The board is compatible with Luxonis’ earlier DepthAI ecosystem and offers many of the same features including object detection, and object tracking. They also offer various pre-trained models such as COVID-19 mask/no-mask detection, age recognition, emotion recognition, face detection, facial landmark, general object detection, pedestrian, and vehicle detection. Myriad X VPU also acts as a real-time video encoder that compresses 4K H.265 video at 30 frames per second. You’ll find the documentation, Python software, and hardware files released under an MIT license in DepthAI repositories on Luxonis Github account.
They tested MobileNetSSD on Raspberry Pi 3B+ with CPU inference at around 6fps, with Intel Neural Compute Stick 2 (NCS2) and OpenVINO at 8.3 fps, and somehow managed to get up to 25.5 fps with megaAI with OpenVINO despite using the same VPU, apparently because they somehow managed to leverage other hardware blocks within Movidius Myriad X, but it’s quite possible the NCS2 data is not directly comparable…
megaAI board has launched on Crowd Supply a little while ago, but we only noticed it as the campaign is about to end in 6 days. The $15,000 funding target has not been reached yet, but with close to $12,000 raised at the time of writing they are not so far from reaching their goal. Rewards start at $169 for one megaAI board with free shipping to the US and $12 to the rest of the world. The $199 RoadRunner megaAI perks should get you a board shipped in early June, but other rewards are scheduled to be shipped at the very end of July 2020.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.