File size: 3,410 Bytes
4dca198 | 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 | """Single-image FFA-Net ONNX inference with side-by-side comparison output."""
import argparse
import sys
from pathlib import Path
import numpy as np
import onnxruntime as ort
from PIL import Image, ImageDraw, ImageFont
FILE = Path(__file__).resolve()
NET_DIR = FILE.parent
ROOT_DIR = NET_DIR.parent
sys.path.insert(0, str(NET_DIR))
def parse_args():
parser = argparse.ArgumentParser(description="FFA-Net ONNX single-image inference.")
parser.add_argument("--onnx", default='onnx/ffa_ots_512x512.onnx', help="ONNX model path.")
parser.add_argument("--input", default='outdoor_natural/nh(2).jpg', help="Path to input hazy image.")
parser.add_argument("--output", default="onnx_compare.png", help="Output comparison image path (hazy | dehazed).")
parser.add_argument("--height", type=int, default=512, help="ONNX input height.")
parser.add_argument("--width", type=int, default=512, help="ONNX input width.")
parser.add_argument("--no_label", action="store_true", help="Do not draw hazy/dehazed labels.")
return parser.parse_args()
MEAN = np.array([0.64, 0.6, 0.58], dtype=np.float32).reshape(3, 1, 1)
STD = np.array([0.14, 0.15, 0.152], dtype=np.float32).reshape(3, 1, 1)
def preprocess(image_path, height, width):
image = Image.open(image_path).convert("RGB")
image = image.resize((width, height), Image.BICUBIC)
arr = np.asarray(image).astype(np.float32) / 255.0
arr = arr.transpose(2, 0, 1)
arr = (arr - MEAN) / STD # 训练同款归一化
return arr[None, ...].astype(np.float32)
def postprocess(output):
arr = np.squeeze(output, axis=0).transpose(1, 2, 0)
arr = np.clip(arr, 0.0, 1.0)
return Image.fromarray((arr * 255.0 + 0.5).astype(np.uint8))
def draw_label(img, text):
draw = ImageDraw.Draw(img)
try:
font = ImageFont.truetype("DejaVuSans-Bold.ttf", max(16, img.height // 40))
except Exception:
font = ImageFont.load_default()
padding = max(5, img.height // 140)
bbox = draw.textbbox((0, 0), text, font=font)
box_w = bbox[2] - bbox[0] + padding * 2
box_h = bbox[3] - bbox[1] + padding * 2
draw.rectangle([0, 0, box_w, box_h], fill=(0, 0, 0))
draw.text((padding, padding), text, fill=(255, 255, 255), font=font)
def make_compare(hazy, dehazed, with_label=True):
hazy = hazy.convert("RGB")
dehazed = dehazed.convert("RGB")
if with_label:
hazy = hazy.copy()
dehazed = dehazed.copy()
draw_label(hazy, "hazy")
draw_label(dehazed, "dehazed")
canvas = Image.new("RGB", (hazy.width + dehazed.width, hazy.height), color=(255, 255, 255))
canvas.paste(hazy, (0, 0))
canvas.paste(dehazed, (hazy.width, 0))
return canvas
def main():
args = parse_args()
hazy_img = Image.open(args.input).convert("RGB").resize((args.width, args.height), Image.BICUBIC)
inp = preprocess(args.input, args.height, args.width)
session = ort.InferenceSession(args.onnx, providers=["CPUExecutionProvider"])
input_name = session.get_inputs()[0].name
out = session.run(None, {input_name: inp})[0]
output_path = Path(args.output)
output_path.parent.mkdir(parents=True, exist_ok=True)
dehazed = postprocess(out)
compare = make_compare(hazy_img, dehazed, with_label=not args.no_label)
compare.save(str(output_path))
print(f"Saved: {output_path}")
if __name__ == "__main__":
main()
|