weedscan-kz / inference_server.py
Rmano-Rudeson's picture
Release recovered WeedScan KZ models, source, and experiment evidence
3eba919 verified
Raw History Blame Contribute Delete
2.4 kB
"""Local FastAPI adapter for the recovered WeedScan KZ frontend."""
import base64
import binascii
import io
import os
from contextlib import asynccontextmanager
from fastapi import FastAPI, HTTPException, Request
from PIL import Image, UnidentifiedImageError
from pydantic import BaseModel, Field
from predict import load_model
@asynccontextmanager
async def lifespan(app):
app.state.model = load_model(os.environ.get("WEEDSCAN_VARIANT", "yolov26n-10class"))
yield
app = FastAPI(title="WeedScan KZ", lifespan=lifespan)
class InferRequest(BaseModel):
imageBase64: str = Field(max_length=20_000_000)
@app.post("/classify")
def classify(payload: InferRequest, request: Request):
try:
raw = base64.b64decode(payload.imageBase64, validate=True)
image = Image.open(io.BytesIO(raw))
if image.width * image.height > 25_000_000:
raise ValueError("Image exceeds 25 megapixels")
image = image.convert("RGB")
except (binascii.Error, UnidentifiedImageError, OSError, ValueError, Image.DecompressionBombError) as exc:
raise HTTPException(status_code=400, detail="Invalid image or unsupported image size") from exc
model = request.app.state.model
result = model.predict(image, device="cpu", verbose=False)[0]
width, height = image.size
boxes, counts = [], {}
weed_count, weed_box_area = 0, 0.0
for box in result.boxes:
x1, y1, x2, y2 = box.xyxy[0].tolist()
label = model.names[int(box.cls[0])]
item = {"x": x1 / width, "y": y1 / height,
"width": (x2 - x1) / width, "height": (y2 - y1) / height,
"label": label, "confidence": round(float(box.conf[0]), 4)}
boxes.append(item)
counts[label] = counts.get(label, 0) + 1
if label.startswith("weed:"):
weed_count += 1
weed_box_area += item["width"] * item["height"]
total = len(boxes)
species = [{"name": name, "count": count, "percentage": round(100 * count / total, 1)}
for name, count in counts.items()]
return {"weedCount": weed_count, "infestationRate": round(min(100 * weed_box_area, 100), 1),
"species": species, "boundingBoxes": boxes,
"summary": f"Detected {weed_count} weed instances among {total} plant detections. "
"Coverage is an uncalibrated bounding-box area estimate."}