car_core/issues.py

from typing import Dict, List
import numpy as np
from PIL import Image, ImageFilter

ISSUE_LIST = [
    "scratch_dent","paint_damage","cracked_windshield","flat_tire","engine_leak",
    "brake_wear","headlight_fault","battery_corrosion","rust","bumper_damage"
]

def _edge_contrast(im: Image.Image) -> float:
    g = im.convert("L").resize((256,256))
    arr = np.array(g, dtype=np.float32)/255.0
    return float(arr.std())

def _redness(im: Image.Image) -> float:
    rgb = im.convert("RGB").resize((256,256))
    arr = np.array(rgb, dtype=np.float32)/255.0
    return float(arr[:,:,0].mean())

def _darkness(im: Image.Image) -> float:
    g = im.convert("L").resize((256,256))
    arr = np.array(g, dtype=np.float32)/255.0
    return float(1.0 - arr.mean())

def detect_issues(images: List[Image.Image]) -> List[str]:
    """Deterministic heuristics over multiple images; choose final set without probabilities."""
    if not images: return ["diagnostic_inspection"]
    # aggregate signals
    contrast = np.mean([_edge_contrast(i) for i in images])
    red = np.mean([_redness(i) for i in images])
    dark = np.mean([_darkness(i) for i in images])
    # thresholds chosen empirically; tune with dataset
    issues = set()
    if contrast > 0.22:
        issues.update(["scratch_dent","paint_damage","bumper_damage"])
    if red > 0.55:
        issues.add("engine_leak")
    if dark > 0.55:
        issues.add("headlight_fault")
    # Always narrow to a final deterministic list of max 4 using rule priority
    priority = ["engine_leak","cracked_windshield","brake_wear","flat_tire","scratch_dent","paint_damage","bumper_damage","headlight_fault","battery_corrosion","rust"]
    final = []
    for p in priority:
        if p in issues:
            final.append(p)
        if len(final) >= 4:
            break
    if not final:
        final = ["diagnostic_inspection"]
    return final