Spaces:
Running
Running
Download index.html from Marconcini/leafscan: direct link, hf CLI and curl.
- Browser
- Download file 8.92 kB
-
https://huggingface.co/spaces/Marconcini/leafscan/resolve/main/index.html
- Command line
-
hf download hf://spaces/Marconcini/leafscan/index.html
-
curl -L -o index.html https://huggingface.co/spaces/Marconcini/leafscan/resolve/main/index.html
8.92 kB
| <html lang="en"> | |
| <head> | |
| <meta charset="utf-8"> | |
| <meta name="viewport" content="width=device-width, initial-scale=1"> | |
| <title>LeafScan</title> | |
| <style> | |
| :root { | |
| --bg: #f7f8f6; --card: #ffffff; --ink: #1f2328; --muted: #59636e; --line: #e3e6e1; | |
| --accent: #2f7d4f; --accent-soft: #e4f1e8; --warn: #9a5b00; --warn-soft: #fbf0dc; | |
| } | |
| @media (prefers-color-scheme: dark) { | |
| :root { --bg: #111412; --card: #1a1f1c; --ink: #e8ece9; --muted: #9aa49d; --line: #2c332e; | |
| --accent: #6cc28e; --accent-soft: #1f3327; --warn: #f0b860; --warn-soft: #3a2e17; } | |
| } | |
| * { box-sizing: border-box; } | |
| body { margin: 0; background: var(--bg); color: var(--ink); | |
| font: 16px/1.5 system-ui, -apple-system, "Segoe UI", sans-serif; } | |
| main { max-width: 860px; margin: 0 auto; padding: 32px 16px 48px; } | |
| h1 { font-size: 1.6rem; margin: 0 0 4px; } | |
| .sub { color: var(--muted); margin: 0 0 24px; } | |
| .grid { display: grid; grid-template-columns: 1fr 1fr; gap: 16px; } | |
| @media (max-width: 680px) { .grid { grid-template-columns: 1fr; } } | |
| .card { background: var(--card); border: 1px solid var(--line); border-radius: 12px; padding: 16px; } | |
| label.drop { display: flex; flex-direction: column; align-items: center; justify-content: center; | |
| min-height: 260px; border: 2px dashed var(--line); border-radius: 10px; cursor: pointer; | |
| color: var(--muted); text-align: center; padding: 12px; overflow: hidden; } | |
| label.drop:hover { border-color: var(--accent); } | |
| label.drop img { max-width: 100%; max-height: 320px; border-radius: 8px; } | |
| input[type=file] { display: none; } | |
| .verdict { font-size: 1.15rem; margin: 0 0 12px; min-height: 1.5em; } | |
| .verdict.unsure { background: var(--warn-soft); color: var(--warn); padding: 10px 12px; border-radius: 8px; font-size: 1rem; } | |
| .bar { margin: 10px 0; } | |
| .bar .name { display: flex; justify-content: space-between; font-size: .92rem; } | |
| .bar .track { height: 8px; background: var(--accent-soft); border-radius: 4px; margin-top: 4px; } | |
| .bar .fill { height: 8px; background: var(--accent); border-radius: 4px; width: 0; transition: width .3s; } | |
| .status { color: var(--muted); font-size: .9rem; } | |
| .about { margin-top: 24px; color: var(--muted); font-size: .92rem; } | |
| .about strong { color: var(--ink); } | |
| a { color: var(--accent); } | |
| </style> | |
| </head> | |
| <body> | |
| <main> | |
| <h1>🌿 LeafScan</h1> | |
| <p class="sub">Plant disease from a leaf photo — runs entirely in your browser, nothing is uploaded.</p> | |
| <div class="grid"> | |
| <div class="card"> | |
| <label class="drop" id="drop"> | |
| <span id="dropText">Click or drop a leaf photo here</span> | |
| <input type="file" id="file" accept="image/*"> | |
| </label> | |
| </div> | |
| <div class="card"> | |
| <p class="verdict" id="verdict"></p> | |
| <div id="bars"></div> | |
| <p class="status" id="status">Loading the model (≈ 32 MB, once)…</p> | |
| </div> | |
| </div> | |
| <div class="about"> | |
| <p><strong>What this is.</strong> A ConvNeXt-T trained on lab photos (PlantVillage), then | |
| fine-tuned on web and field photos (PlantDoc), for <strong>38 crop / disease classes</strong> | |
| of 14 crops. This page runs an int8-quantised copy of the model with ONNX Runtime Web.</p> | |
| <p><strong>How much to trust it.</strong> On held-out field photos it is right about | |
| <strong>72%</strong> of the time. Below a confidence threshold fitted on field photos it says | |
| "Not sure": it answers about two thirds of field photos, and about <strong>85%</strong> of those | |
| answers are right. It only knows these 38 classes and is <strong>not</strong> a substitute for | |
| an agronomist.</p> | |
| <p><strong>Beyond this page.</strong> The full project adds an assistant that answers your | |
| questions about the diagnosed disease from 28 Wikipedia articles, with a local language model | |
| that cites its sources, says "I don't know" when they do not answer, and flags sentences its | |
| sources do not support. Its models are about 10 GB, so it runs on your own computer | |
| (instructions in the repository).</p> | |
| <p>Code, experiments and limitations: | |
| <a href="https://github.com/marconcini/leafscan">github.com/marconcini/leafscan</a> · | |
| model: <a href="https://huggingface.co/Marconcini/leafscan-convnext-tiny">Marconcini/leafscan-convnext-tiny</a></p> | |
| </div> | |
| </main> | |
| <script src="https://cdn.jsdelivr.net/npm/onnxruntime-web@1.20.1/dist/ort.min.js"></script> | |
| <script> | |
| // Model files: the Hugging Face model repo by default, or ?base=<url> (local testing). | |
| const BASE = new URLSearchParams(location.search).get("base") | |
| || "https://huggingface.co/Marconcini/leafscan-convnext-tiny/resolve/main/"; | |
| const MEAN = [0.485, 0.456, 0.406], STD = [0.229, 0.224, 0.225]; | |
| ort.env.wasm.wasmPaths = "https://cdn.jsdelivr.net/npm/onnxruntime-web@1.20.1/dist/"; | |
| let session, config; | |
| const $ = (id) => document.getElementById(id); | |
| function prettyLabel(label) { // same as leafscan.data.pretty_label | |
| const [crop, disease] = label.split("___"); | |
| return `${crop.replaceAll("_", " ").replaceAll(",", "").trim()} — ${disease.replaceAll("_", " ").trim()}`; | |
| } | |
| // Same geometry as the evaluation transform: shorter side -> round(1.14 * size), centre crop. | |
| function preprocess(img, size) { | |
| const short = Math.floor(size * 1.14); | |
| const scale = short / Math.min(img.naturalWidth, img.naturalHeight); | |
| const w = Math.round(img.naturalWidth * scale), h = Math.round(img.naturalHeight * scale); | |
| const canvas = document.createElement("canvas"); | |
| canvas.width = size; canvas.height = size; | |
| const ctx = canvas.getContext("2d", { willReadFrequently: true }); | |
| ctx.imageSmoothingEnabled = true; ctx.imageSmoothingQuality = "high"; | |
| ctx.drawImage(img, -Math.round((w - size) / 2), -Math.round((h - size) / 2), w, h); | |
| const px = ctx.getImageData(0, 0, size, size).data; | |
| const data = new Float32Array(3 * size * size); | |
| for (let i = 0; i < size * size; i++) | |
| for (let c = 0; c < 3; c++) | |
| data[c * size * size + i] = (px[4 * i + c] / 255 - MEAN[c]) / STD[c]; | |
| return new ort.Tensor("float32", data, [1, 3, size, size]); | |
| } | |
| async function predict(img) { | |
| const { file, temperature, threshold } = config.onnx_int8; | |
| const out = await session.run({ image: preprocess(img, config.img_size) }); | |
| const logits = Array.from(out.logits.data).map((z) => z / temperature); | |
| const m = Math.max(...logits); | |
| const exps = logits.map((z) => Math.exp(z - m)); | |
| const sum = exps.reduce((a, b) => a + b, 0); | |
| const probs = exps.map((e) => e / sum); | |
| const top = probs.map((p, i) => [config.classes[i], p]).sort((a, b) => b[1] - a[1]).slice(0, 3); | |
| return { top, confident: top[0][1] >= threshold, threshold }; | |
| } | |
| function show(pred) { | |
| const [name, p] = pred.top[0]; | |
| const v = $("verdict"); | |
| if (pred.confident) { | |
| v.className = "verdict"; | |
| v.innerHTML = `<strong>${prettyLabel(name)}</strong> (confidence ${Math.round(100 * p)}%)`; | |
| } else { | |
| v.className = "verdict unsure"; | |
| v.innerHTML = `<strong>Not sure.</strong> Best guess: ${prettyLabel(name)} (${Math.round(100 * p)}%), | |
| below the ${Math.round(100 * pred.threshold)}% confidence needed to answer. | |
| Try a closer photo of a single leaf, or ask an expert.`; | |
| } | |
| $("bars").innerHTML = pred.top.map(([c, q]) => ` | |
| <div class="bar"><div class="name"><span>${prettyLabel(c)}</span><span>${Math.round(100 * q)}%</span></div> | |
| <div class="track"><div class="fill" style="width:${(100 * q).toFixed(1)}%"></div></div></div>`).join(""); | |
| } | |
| async function handleFile(file) { | |
| if (!file || !session) return; | |
| const img = new Image(); | |
| img.src = URL.createObjectURL(file); | |
| await img.decode(); | |
| $("dropText").replaceWith(Object.assign(img, { id: "dropText" })); | |
| $("status").textContent = "Thinking…"; | |
| const t0 = performance.now(); | |
| show(await predict(img)); | |
| $("status").textContent = `Done in ${Math.round(performance.now() - t0)} ms, on your device.`; | |
| } | |
| // Test hook: predict from an image URL (used to compare with the Python implementation). | |
| window.leafscanPredictUrl = async (url) => { | |
| const img = new Image(); img.crossOrigin = "anonymous"; img.src = url; await img.decode(); | |
| return predict(img); | |
| }; | |
| (async () => { | |
| try { | |
| config = await (await fetch(BASE + "config.json")).json(); | |
| session = await ort.InferenceSession.create(BASE + config.onnx_int8.file, | |
| { executionProviders: ["wasm"] }); | |
| $("status").textContent = "Ready. Choose a photo."; | |
| window.leafscanReady = true; | |
| } catch (e) { | |
| $("status").textContent = "Could not load the model: " + e; | |
| window.leafscanError = String(e); | |
| } | |
| })(); | |
| $("file").addEventListener("change", (e) => handleFile(e.target.files[0])); | |
| const drop = $("drop"); | |
| drop.addEventListener("dragover", (e) => e.preventDefault()); | |
| drop.addEventListener("drop", (e) => { e.preventDefault(); handleFile(e.dataTransfer.files[0]); }); | |
| </script> | |
| </body> | |
| </html> | |