File size: 7,759 Bytes
39594de ca6e516 39594de ca6e516 39594de 93f67bb 39594de 80d8a8b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 |
import io
import os
import gdown
import base64
from typing import Optional
import cv2
import numpy as np
from PIL import Image
from fastapi import FastAPI, UploadFile, File, Form
from fastapi.responses import JSONResponse
from fastapi.middleware.cors import CORSMiddleware
from detectron2.engine import DefaultPredictor
from detectron2.config import get_cfg
from detectron2.projects.point_rend import add_pointrend_config
# -------------------------------
# FastAPI setup
# -------------------------------
app = FastAPI(title="Rooftop Segmentation API")
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
# -------------------------------
# Available epsilons
# -------------------------------
EPSILONS = [0.01, 0.005, 0.004, 0.003, 0.001]
@app.get("/epsilons")
def get_epsilons():
return {"epsilons": EPSILONS}
# -------------------------------
# Google Drive model download (irregular-flat)
# -------------------------------
MODEL_PATH_IRREGULAR = "/tmp/model_irregular_flat.pth"
DRIVE_FILE_ID = "15vi4zPhCs3aBnGepVnXFOqQjxdK1jpnA"
def download_irregular_model():
if not os.path.exists(MODEL_PATH_IRREGULAR):
url = f"https://drive.google.com/uc?id={DRIVE_FILE_ID}"
tmp_dir = "/tmp/gdown"
os.makedirs(tmp_dir, exist_ok=True)
os.environ["GDOWN_CACHE_DIR"] = tmp_dir
print("Downloading irregular-flat Detectron2 model...")
gdown.download(url, MODEL_PATH_IRREGULAR, quiet=False, fuzzy=True, use_cookies=False)
print("Download complete.")
else:
print("Irregular-flat model already exists, skipping download.")
download_irregular_model()
if os.path.exists(MODEL_PATH_IRREGULAR):
print("Irregular-flat model is ready at", MODEL_PATH_IRREGULAR)
else:
print("Irregular-flat model NOT found! Something went wrong!")
# -------------------------------
# Detectron2 model setup
# -------------------------------
def setup_model_rect(weights_path: str):
cfg = get_cfg()
add_pointrend_config(cfg)
cfg_path = "detectron2/projects/PointRend/configs/InstanceSegmentation/pointrend_rcnn_R_50_FPN_3x_coco.yaml"
cfg.merge_from_file(cfg_path)
cfg.MODEL.ROI_HEADS.NUM_CLASSES = 2
cfg.MODEL.POINT_HEAD.NUM_CLASSES = cfg.MODEL.ROI_HEADS.NUM_CLASSES
cfg.MODEL.WEIGHTS = weights_path
cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = 0.5
cfg.MODEL.DEVICE = "cpu"
return DefaultPredictor(cfg)
def setup_model_irregular(weights_path: str):
cfg = get_cfg()
add_pointrend_config(cfg)
cfg_path = "detectron2/projects/PointRend/configs/InstanceSegmentation/pointrend_rcnn_R_50_FPN_3x_coco.yaml"
cfg.merge_from_file(cfg_path)
cfg.MODEL.ROI_HEADS.NUM_CLASSES = 1
cfg.MODEL.POINT_HEAD.NUM_CLASSES = cfg.MODEL.ROI_HEADS.NUM_CLASSES
cfg.MODEL.WEIGHTS = weights_path
cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = 0.5
cfg.MODEL.DEVICE = "cpu"
return DefaultPredictor(cfg)
# Load models
predictor_rect = setup_model_rect("/app/model_rect_final.pth")
predictor_irregular_flat = setup_model_irregular(MODEL_PATH_IRREGULAR)
# -------------------------------
# Post-processing functions
# -------------------------------
def postprocess_rect(mask: np.ndarray, epsilon: float) -> Optional[np.ndarray]:
mask_uint8 = (mask * 255).astype(np.uint8)
contours, _ = cv2.findContours(mask_uint8, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
if not contours:
return None
c = max(contours, key=cv2.contourArea)
eps = epsilon * cv2.arcLength(c, True)
approx = cv2.approxPolyDP(c, eps, True)
simp = np.zeros_like(mask_uint8)
cv2.fillPoly(simp, [approx], 255)
return simp
def postprocess_irregular(mask: np.ndarray, epsilon: float) -> Optional[np.ndarray]:
mask_uint8 = (mask * 255).astype(np.uint8)
contours, _ = cv2.findContours(mask_uint8, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
if not contours:
return None
c = max(contours, key=cv2.contourArea)
eps = epsilon * cv2.arcLength(c, True)
polygon = cv2.approxPolyDP(c, eps, True)
return polygon.reshape(-1, 2)
def mask_to_polygon(mask: np.ndarray) -> Optional[np.ndarray]:
contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
if not contours:
return None
largest = max(contours, key=cv2.contourArea)
return largest.reshape(-1, 2)
def extract_polygon_vertices(mask: np.ndarray, epsilon_ratio: float = 0.004):
"""
Extract clean polygon vertices from a binary mask.
Returns Nx2 array of vertices.
"""
mask_uint8 = (mask * 255).astype(np.uint8)
contours, _ = cv2.findContours(mask_uint8, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
if not contours:
return None
c = max(contours, key=cv2.contourArea)
epsilon = epsilon_ratio * cv2.arcLength(c, True)
polygon = cv2.approxPolyDP(c, epsilon, True)
return polygon.reshape(-1, 2)
def im_to_b64_png(im: np.ndarray) -> str:
_, buffer = cv2.imencode(".png", im)
return base64.b64encode(buffer).decode()
def overlay_polygon(im: np.ndarray, polygon: Optional[np.ndarray]) -> np.ndarray:
overlay = im.copy()
if polygon is not None:
# Draw polygon outline
cv2.polylines(overlay, [polygon.astype(np.int32)], True, (0, 255, 0), 2)
# Draw vertex points (red circles)
for (x, y) in polygon:
cv2.circle(overlay, (int(x), int(y)), radius=4, color=(0, 0, 255), thickness=-1)
return overlay
# -------------------------------
# API endpoints
# -------------------------------
@app.get("/")
def root():
return {"message": "Rooftop Segmentation API is running!"}
@app.post("/predict")
async def predict(
file: UploadFile = File(...),
rooftop_type: str = Form(...),
epsilon: float = Form(0.004)
):
contents = await file.read()
try:
im_pil = Image.open(io.BytesIO(contents)).convert("RGB")
except Exception as e:
return JSONResponse(status_code=400, content={"error": "Invalid image", "detail": str(e)})
im = np.array(im_pil)[:, :, ::-1].copy() # RGB -> BGR
if rooftop_type.lower() == "rectangular":
predictor = predictor_rect
post_fn = lambda mask: postprocess_rect(mask, epsilon)
model_used = "model_rect_final.pth"
elif rooftop_type.lower() == "irregular":
predictor = predictor_irregular_flat
post_fn = lambda mask: postprocess_irregular(mask, epsilon)
model_used = "model_irregular_flat.pth"
else:
return JSONResponse(status_code=400, content={"error": "Invalid rooftop_type. Choose 'rectangular' or 'irregular'."})
outputs = predictor(im)
instances = outputs["instances"].to("cpu")
if len(instances) == 0:
return {
"polygon": None,
"vertices": None,
"vertex_count": 0,
"image": None,
"model_used": model_used,
"rooftop_type": rooftop_type,
"epsilon": epsilon
}
idx = int(instances.scores.argmax().item())
raw_mask = instances.pred_masks[idx].numpy().astype(np.uint8)
result_mask = post_fn(raw_mask)
polygon = mask_to_polygon(result_mask) if rooftop_type.lower() == "rectangular" else result_mask
# --- Vertices extraction ---
# vertices = extract_polygon_vertices(raw_mask, epsilon)
# vertex_count = len(vertices) if vertices is not None else 0
overlay = overlay_polygon(im, polygon)
img_b64 = im_to_b64_png(overlay)
return {
"polygon": polygon.tolist() if polygon is not None else None,
"image": img_b64,
"model_used": model_used,
"rooftop_type": rooftop_type,
"epsilon": epsilon
}
|