{"slug": "avion-0-1-a-zero-allocation-perception-front-end-for-robots-and-ai-in-one-c99", "title": "AVION 0.1: a zero-allocation perception front end for robots and AI, in one C99 header", "summary": "AVION 0.1, released as a single C99 header and published on Hugging Face, is a zero-allocation perception front end that runs in front of a robot's vision model on the GPU and hands the network a foveated NCHW tensor in fp32 or fp16 with no copy. The release reports 0 heap allocations per 1,000 frames versus 132,292 for ffmpeg and 20,000 for OpenCV DIS on the same machine and allocation counter, and derives a virtual IMU accurate to 0.031 px rms, a stabilized event-camera stream that cuts a shaken clip's 17.2M events to 1.71M, time to contact of 1.24 s against a true 1.33 s, a 71.98 BPM pulse from camera shake (72.00 from Cr chroma alone), and scene cuts from H.264/HEVC slice headers in 0.23 µs per 4K frame. The author states where the approach loses, noting VLC plays 720p with less CPU.", "body_md": "AVION 0.1: a perception front end for robot brains, in one C99 header\n\nMost vision stacks feed a model frames that were decoded, copied and allocated for. AVION does the work in front of the model on the GPU, with zero heap allocations while it runs, and hands the network a foveated NCHW tensor (fp32/fp16) with no copy.\n\nWhat it gives a model, from an ordinary camera:\n\n• A virtual IMU: the camera’s shake read from pixels, 0.031 px rms off the truth\n\n• A stabilised event-camera (DVS) stream: a shaken clip’s 17.2M events become 1.71M, only what moves\n\n• Time to contact: 1.24 s read for a true 1.33 s\n\n• A pulse through camera shake: 71.98 BPM; 72.00 from Cr chroma alone\n\n• Scene cuts from H.264/HEVC slice headers before decoding: 0.23 µs a 4K frame\n\nHead to head on one machine, one allocation counter: 0 allocations per 1,000 frames against 132,292 (ffmpeg) and 20,000 (OpenCV DIS). And where we lose, we say so: VLC plays 720p with less CPU.\n\nEvery picture is a snapshot the engine wrote itself, against a truth we set.\n\nModel card: [alexandre5645/avion-0.1 · Hugging Face](https://huggingface.co/alexandre5645/avion-0.1)\n\nSite, method and evaluation bundle: [AVION 0.1](https://5645hm-a11y.github.io/avion/)", "url": "https://wpnews.pro/news/avion-0-1-a-zero-allocation-perception-front-end-for-robots-and-ai-in-one-c99", "canonical_source": "https://discuss.huggingface.co/t/avion-0-1-a-zero-allocation-perception-front-end-for-robots-and-ai-in-one-c99-header/191403#post_1", "published_at": "2026-11-03 18:15:13+00:00", "updated_at": "2026-11-03 18:19:37.907306+00:00", "lang": "en", "topics": ["computer-vision", "robotics", "ai-infrastructure", "developer-tools", "ai-tools"], "entities": ["AVION 0.1", "Hugging Face", "ffmpeg", "OpenCV DIS", "VLC"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/avion-0-1-a-zero-allocation-perception-front-end-for-robots-and-ai-in-one-c99", "markdown": "https://wpnews.pro/news/avion-0-1-a-zero-allocation-perception-front-end-for-robots-and-ai-in-one-c99.md", "text": "https://wpnews.pro/news/avion-0-1-a-zero-allocation-perception-front-end-for-robots-and-ai-in-one-c99.txt", "jsonld": "https://wpnews.pro/news/avion-0-1-a-zero-allocation-perception-front-end-for-robots-and-ai-in-one-c99.jsonld"}}