Initial Bioportal Lab prototype

server.py

@@ -0,0 +1,61 @@
+from __future__ import annotations
+
+import io
+from pathlib import Path
+
+import joblib
+import numpy as np
+from flask import Flask, jsonify, request, send_from_directory
+
+from petri_cv import candidate_points, features_for_points, prepare_image
+
+ROOT = Path(__file__).resolve().parent
+MODEL_PATH = ROOT / "models/petri_candidate_classifier.joblib"
+app = Flask(__name__)
+model_bundle = joblib.load(MODEL_PATH) if MODEL_PATH.exists() else None
+
+
+@app.get("/")
+def index():
+    return send_from_directory(ROOT, "index.html")
+
+
+@app.get("/<path:path>")
+def assets(path: str):
+    return send_from_directory(ROOT, path)
+
+
+@app.post("/api/analyze")
+def analyze():
+    if model_bundle is None:
+        return jsonify(error="Модель не обучена. Запустите train_petri_model.py."), 503
+    uploaded = request.files.get("image")
+    if uploaded is None or not uploaded.mimetype.startswith("image/"):
+        return jsonify(error="Передайте изображение PNG, JPG или WEBP."), 400
+    try:
+        prepared = prepare_image(io.BytesIO(uploaded.read()))
+    except Exception:
+        return jsonify(error="Не удалось прочитать изображение."), 400
+    points = candidate_points(prepared)
+    if not points:
+        return jsonify(colonies=[], candidates=0, model="RandomForest candidate ranker")
+    probabilities = model_bundle["model"].predict_proba(features_for_points(prepared, points))[:, 1]
+    ranked = sorted(zip(probabilities, points), reverse=True)
+    selected = []
+    # Non-maximum suppression avoids several markers on a single colony.
+    for score, (x, y) in ranked:
+        if score < .54:
+            continue
+        if all((x - candidate["x_px"]) ** 2 + (y - candidate["y_px"]) ** 2 > 36 ** 2 for candidate in selected):
+            selected.append({"x": round(x / prepared.gray.shape[1], 5), "y": round(y / prepared.gray.shape[0], 5),
+                             "x_px": x, "y_px": y, "score": round(float(score), 3)})
+        if len(selected) >= 300:
+            break
+    for item in selected:
+        item.pop("x_px")
+        item.pop("y_px")
+    return jsonify(colonies=selected, candidates=len(points), model="RandomForest candidate ranker")
+
+
+if __name__ == "__main__":
+    app.run(host="127.0.0.1", port=8000, debug=False)