SafeCamp
Point your phone at a plant, insect, nest or patch of ground and get a four-level campsite safety verdict — fully offline, nothing uploaded.
项目介绍
SafeCamp is an offline campsite safety inspector that works with no signal. Point your phone at a plant, an insect, a nest or a patch of ground and it tells you how worried to be and what to do next, in four levels: Favorable, Caution, Danger and Critical. One example of its advice: move your tent pitch at least 30 m away. The whole check runs on the phone, so a photo of your campsite never leaves the device.
The checker is CLIP (Xenova/clip-vit-base-patch32, int8 ONNX) through Transformers.js, running on WebAssembly inside a Web Worker so the interface never freezes. There is no training: CLIP compares the photo with text prompts such as a photo of a paper wasp nest with hexagonal honeycomb cells, and the closest prompts win. The hazard list has 41 entries in one TypeScript file: 37 hazards covering poisonous plants, stinging insects, snakes and large animals, unsafe terrain and water, and fire and weather risks, plus 4 favorable signs such as a flat, dry tent site. Because the text side never changes, the text embeddings are encoded once at dev time, bringing a scan back under 1.5 seconds.
The app is built to avoid a wrong reassurance. An empty bird nest once came back as wasp nest at 97 percent because there was no bird nest class, so 25 everyday labels (grass, bark, bird nests, rocks) were added so ordinary things stop being forced into a hazard. A real hazard always outranks a favorable sign, related hazards pool their scores, low confidence gives a possible, be careful warning that names the runner-up, and no known hazard matched is never shown as safe. A Why panel shows the top scores and says plainly that the score is a relative share between the app's own labels, not a probability. Vibe-coded evidence, in the author's own words on DEV (11 October 2026): "I built SafeCamp with Claude as a pair programmer for design discussion and code review. I ran, tested and debugged everything myself." Note: the author says Claude, not Claude Code, so aiTools is recorded as Claude only and is not upgraded. GitHub stars: 0 at the 28 October 2026 check.
Honest limits, as the author states them: SafeCamp is a first-pass aid, not an authority. CLIP is a general model and will be wrong sometimes, especially on blurry, dark, distant or unusual photos. Mushrooms, berries and snakes always say treat as dangerous, because a general vision model cannot reliably tell harmless ones from deadly ones. It has not been tested on a wide range of phones, hazards differ by region and the advice is general: follow local rangers, signs and emergency services. The demo runs on Render's free tier, so the first load after a quiet spell can take a while to wake up.