Spaces:
Running
Running
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,578 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from __future__ import annotations
|
2 |
+
import os
|
3 |
+
import gc
|
4 |
+
import base64
|
5 |
+
import io
|
6 |
+
import time
|
7 |
+
import shutil
|
8 |
+
import numpy as np
|
9 |
+
import torch
|
10 |
+
import cv2
|
11 |
+
import ezdxf
|
12 |
+
import gradio as gr
|
13 |
+
from PIL import Image, ImageEnhance
|
14 |
+
from pathlib import Path
|
15 |
+
from typing import List, Union
|
16 |
+
from ultralytics import YOLOWorld, YOLO
|
17 |
+
from ultralytics.engine.results import Results
|
18 |
+
from ultralytics.utils.plotting import save_one_box
|
19 |
+
from transformers import AutoModelForImageSegmentation
|
20 |
+
from torchvision import transforms
|
21 |
+
from scalingtestupdated import calculate_scaling_factor
|
22 |
+
from shapely.geometry import Polygon, Point, MultiPolygon
|
23 |
+
from scipy.interpolate import splprep, splev
|
24 |
+
from scipy.ndimage import gaussian_filter1d
|
25 |
+
from u2net import U2NETP
|
26 |
+
|
27 |
+
# ---------------------
|
28 |
+
# Create a cache folder for models
|
29 |
+
# ---------------------
|
30 |
+
CACHE_DIR = os.path.join(os.path.dirname(__file__), ".cache")
|
31 |
+
os.makedirs(CACHE_DIR, exist_ok=True)
|
32 |
+
|
33 |
+
# ---------------------
|
34 |
+
# Custom Exceptions
|
35 |
+
# ---------------------
|
36 |
+
class DrawerNotDetectedError(Exception):
|
37 |
+
"""Raised when the drawer cannot be detected in the image"""
|
38 |
+
pass
|
39 |
+
|
40 |
+
class ReferenceBoxNotDetectedError(Exception):
|
41 |
+
"""Raised when the reference box cannot be detected in the image"""
|
42 |
+
pass
|
43 |
+
|
44 |
+
# ---------------------
|
45 |
+
# Global Model Initialization with caching and print statements
|
46 |
+
# ---------------------
|
47 |
+
print("Loading YOLOWorld model...")
|
48 |
+
start_time = time.time()
|
49 |
+
yolo_model_path = os.path.join(CACHE_DIR, "yolov8x-worldv2.pt")
|
50 |
+
if not os.path.exists(yolo_model_path):
|
51 |
+
print("Caching YOLOWorld model to", yolo_model_path)
|
52 |
+
shutil.copy("yolov8x-worldv2.pt", yolo_model_path)
|
53 |
+
drawer_detector_global = YOLOWorld(yolo_model_path)
|
54 |
+
drawer_detector_global.set_classes(["box"])
|
55 |
+
print("YOLOWorld model loaded in {:.2f} seconds".format(time.time() - start_time))
|
56 |
+
|
57 |
+
print("Loading YOLO reference model...")
|
58 |
+
start_time = time.time()
|
59 |
+
reference_model_path = os.path.join(CACHE_DIR, "last.pt")
|
60 |
+
if not os.path.exists(reference_model_path):
|
61 |
+
print("Caching YOLO reference model to", reference_model_path)
|
62 |
+
shutil.copy("last.pt", reference_model_path)
|
63 |
+
reference_detector_global = YOLO(reference_model_path)
|
64 |
+
print("YOLO reference model loaded in {:.2f} seconds".format(time.time() - start_time))
|
65 |
+
|
66 |
+
print("Loading U²-Net model for reference background removal (U2NETP)...")
|
67 |
+
start_time = time.time()
|
68 |
+
u2net_model_path = os.path.join(CACHE_DIR, "u2netp.pth")
|
69 |
+
if not os.path.exists(u2net_model_path):
|
70 |
+
print("Caching U²-Net model to", u2net_model_path)
|
71 |
+
shutil.copy("u2netp.pth", u2net_model_path)
|
72 |
+
u2net_global = U2NETP(3, 1)
|
73 |
+
u2net_global.load_state_dict(torch.load(u2net_model_path, map_location="cpu"))
|
74 |
+
device = "cpu"
|
75 |
+
u2net_global.to(device)
|
76 |
+
u2net_global.eval()
|
77 |
+
print("U²-Net model loaded in {:.2f} seconds".format(time.time() - start_time))
|
78 |
+
|
79 |
+
print("Loading BiRefNet model...")
|
80 |
+
start_time = time.time()
|
81 |
+
birefnet_global = AutoModelForImageSegmentation.from_pretrained(
|
82 |
+
"zhengpeng7/BiRefNet", trust_remote_code=True, cache_dir=CACHE_DIR
|
83 |
+
)
|
84 |
+
torch.set_float32_matmul_precision("high")
|
85 |
+
birefnet_global.to(device)
|
86 |
+
birefnet_global.eval()
|
87 |
+
print("BiRefNet model loaded in {:.2f} seconds".format(time.time() - start_time))
|
88 |
+
|
89 |
+
# Define transform for BiRefNet
|
90 |
+
transform_image_global = transforms.Compose([
|
91 |
+
transforms.Resize((1024, 1024)),
|
92 |
+
transforms.ToTensor(),
|
93 |
+
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
|
94 |
+
])
|
95 |
+
|
96 |
+
# ---------------------
|
97 |
+
# Model Reload Function (if needed)
|
98 |
+
# ---------------------
|
99 |
+
def unload_and_reload_models():
|
100 |
+
global drawer_detector_global, reference_detector_global, birefnet_global, u2net_global
|
101 |
+
print("Reloading models...")
|
102 |
+
start_time = time.time()
|
103 |
+
del drawer_detector_global, reference_detector_global, birefnet_global, u2net_global
|
104 |
+
gc.collect()
|
105 |
+
if torch.cuda.is_available():
|
106 |
+
torch.cuda.empty_cache()
|
107 |
+
gc.collect()
|
108 |
+
new_drawer_detector = YOLOWorld(os.path.join(CACHE_DIR, "yolov8x-worldv2.pt"))
|
109 |
+
new_drawer_detector.set_classes(["box"])
|
110 |
+
new_reference_detector = YOLO(os.path.join(CACHE_DIR, "last.pt"))
|
111 |
+
new_birefnet = AutoModelForImageSegmentation.from_pretrained(
|
112 |
+
"zhengpeng7/BiRefNet", trust_remote_code=True, cache_dir=CACHE_DIR
|
113 |
+
)
|
114 |
+
new_birefnet.to(device)
|
115 |
+
new_birefnet.eval()
|
116 |
+
new_u2net = U2NETP(3, 1)
|
117 |
+
new_u2net.load_state_dict(torch.load(os.path.join(CACHE_DIR, "u2netp.pth"), map_location="cpu"))
|
118 |
+
new_u2net.to(device)
|
119 |
+
new_u2net.eval()
|
120 |
+
drawer_detector_global = new_drawer_detector
|
121 |
+
reference_detector_global = new_reference_detector
|
122 |
+
birefnet_global = new_birefnet
|
123 |
+
u2net_global = new_u2net
|
124 |
+
print("Models reloaded in {:.2f} seconds".format(time.time() - start_time))
|
125 |
+
|
126 |
+
# ---------------------
|
127 |
+
# Helper Function: resize_img (defined once)
|
128 |
+
# ---------------------
|
129 |
+
def resize_img(img: np.ndarray, resize_dim):
|
130 |
+
return np.array(Image.fromarray(img).resize(resize_dim))
|
131 |
+
|
132 |
+
# ---------------------
|
133 |
+
# Other Helper Functions for Detection & Processing
|
134 |
+
# ---------------------
|
135 |
+
def yolo_detect(image: Union[str, Path, int, Image.Image, list, tuple, np.ndarray, torch.Tensor]) -> np.ndarray:
|
136 |
+
t = time.time()
|
137 |
+
results: List[Results] = drawer_detector_global.predict(image)
|
138 |
+
if not results or len(results) == 0 or len(results[0].boxes) == 0:
|
139 |
+
raise DrawerNotDetectedError("Drawer not detected in the image.")
|
140 |
+
print("Drawer detection completed in {:.2f} seconds".format(time.time() - t))
|
141 |
+
return save_one_box(results[0].cpu().boxes.xyxy, im=results[0].orig_img, save=False)
|
142 |
+
|
143 |
+
def detect_reference_square(img: np.ndarray):
|
144 |
+
t = time.time()
|
145 |
+
res = reference_detector_global.predict(img, conf=0.45)
|
146 |
+
if not res or len(res) == 0 or len(res[0].boxes) == 0:
|
147 |
+
raise ReferenceBoxNotDetectedError("Reference box not detected in the image.")
|
148 |
+
print("Reference detection completed in {:.2f} seconds".format(time.time() - t))
|
149 |
+
return (
|
150 |
+
save_one_box(res[0].cpu().boxes.xyxy, res[0].orig_img, save=False),
|
151 |
+
res[0].cpu().boxes.xyxy[0]
|
152 |
+
)
|
153 |
+
|
154 |
+
# Use U2NETP for reference background removal.
|
155 |
+
def remove_bg_u2netp(image: np.ndarray) -> np.ndarray:
|
156 |
+
t = time.time()
|
157 |
+
image_pil = Image.fromarray(image)
|
158 |
+
transform_u2netp = transforms.Compose([
|
159 |
+
transforms.Resize((320, 320)),
|
160 |
+
transforms.ToTensor(),
|
161 |
+
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
|
162 |
+
])
|
163 |
+
input_tensor = transform_u2netp(image_pil).unsqueeze(0).to("cpu")
|
164 |
+
with torch.no_grad():
|
165 |
+
outputs = u2net_global(input_tensor)
|
166 |
+
pred = outputs[0]
|
167 |
+
pred = (pred - pred.min()) / (pred.max() - pred.min() + 1e-8)
|
168 |
+
pred_np = pred.squeeze().cpu().numpy()
|
169 |
+
pred_np = cv2.resize(pred_np, (image_pil.width, image_pil.height))
|
170 |
+
pred_np = (pred_np * 255).astype(np.uint8)
|
171 |
+
print("U2NETP background removal completed in {:.2f} seconds".format(time.time() - t))
|
172 |
+
return pred_np
|
173 |
+
|
174 |
+
# Use BiRefNet for main object background removal.
|
175 |
+
def remove_bg(image: np.ndarray) -> np.ndarray:
|
176 |
+
t = time.time()
|
177 |
+
image_pil = Image.fromarray(image)
|
178 |
+
input_images = transform_image_global(image_pil).unsqueeze(0).to("cpu")
|
179 |
+
with torch.no_grad():
|
180 |
+
preds = birefnet_global(input_images)[-1].sigmoid().cpu()
|
181 |
+
pred = preds[0].squeeze()
|
182 |
+
pred_pil = transforms.ToPILImage()(pred)
|
183 |
+
scale_ratio = 1024 / max(image_pil.size)
|
184 |
+
scaled_size = (int(image_pil.size[0] * scale_ratio), int(image_pil.size[1] * scale_ratio))
|
185 |
+
result = np.array(pred_pil.resize(scaled_size))
|
186 |
+
print("BiRefNet background removal completed in {:.2f} seconds".format(time.time() - t))
|
187 |
+
return result
|
188 |
+
|
189 |
+
def make_square(img: np.ndarray):
|
190 |
+
height, width = img.shape[:2]
|
191 |
+
max_dim = max(height, width)
|
192 |
+
pad_height = (max_dim - height) // 2
|
193 |
+
pad_width = (max_dim - width) // 2
|
194 |
+
pad_height_extra = max_dim - height - 2 * pad_height
|
195 |
+
pad_width_extra = max_dim - width - 2 * pad_width
|
196 |
+
if len(img.shape) == 3:
|
197 |
+
padded = np.pad(img, ((pad_height, pad_height + pad_height_extra),
|
198 |
+
(pad_width, pad_width + pad_width_extra),
|
199 |
+
(0, 0)), mode="edge")
|
200 |
+
else:
|
201 |
+
padded = np.pad(img, ((pad_height, pad_height + pad_height_extra),
|
202 |
+
(pad_width, pad_width + pad_width_extra)), mode="edge")
|
203 |
+
return padded
|
204 |
+
|
205 |
+
def shrink_bbox(image: np.ndarray, shrink_factor: float):
|
206 |
+
height, width = image.shape[:2]
|
207 |
+
center_x, center_y = width // 2, height // 2
|
208 |
+
new_width = int(width * shrink_factor)
|
209 |
+
new_height = int(height * shrink_factor)
|
210 |
+
x1 = max(center_x - new_width // 2, 0)
|
211 |
+
y1 = max(center_y - new_height // 2, 0)
|
212 |
+
x2 = min(center_x + new_width // 2, width)
|
213 |
+
y2 = min(center_y + new_height // 2, height)
|
214 |
+
return image[y1:y2, x1:x2]
|
215 |
+
|
216 |
+
def exclude_scaling_box(image: np.ndarray, bbox: np.ndarray, orig_size: tuple, processed_size: tuple, expansion_factor: float = 1.2) -> np.ndarray:
|
217 |
+
x_min, y_min, x_max, y_max = map(int, bbox)
|
218 |
+
scale_x = processed_size[1] / orig_size[1]
|
219 |
+
scale_y = processed_size[0] / orig_size[0]
|
220 |
+
x_min = int(x_min * scale_x)
|
221 |
+
x_max = int(x_max * scale_x)
|
222 |
+
y_min = int(y_min * scale_y)
|
223 |
+
y_max = int(y_max * scale_y)
|
224 |
+
box_width = x_max - x_min
|
225 |
+
box_height = y_max - y_min
|
226 |
+
expanded_x_min = max(0, int(x_min - (expansion_factor - 1) * box_width / 2))
|
227 |
+
expanded_x_max = min(image.shape[1], int(x_max + (expansion_factor - 1) * box_width / 2))
|
228 |
+
expanded_y_min = max(0, int(y_min - (expansion_factor - 1) * box_height / 2))
|
229 |
+
expanded_y_max = min(image.shape[0], int(y_max + (expansion_factor - 1) * box_height / 2))
|
230 |
+
image[expanded_y_min:expanded_y_max, expanded_x_min:expanded_x_max] = 0
|
231 |
+
return image
|
232 |
+
|
233 |
+
def resample_contour(contour):
|
234 |
+
num_points = 1000
|
235 |
+
smoothing_factor = 5
|
236 |
+
spline_degree = 3
|
237 |
+
if len(contour) < spline_degree + 1:
|
238 |
+
raise ValueError(f"Contour must have at least {spline_degree + 1} points, but has {len(contour)} points.")
|
239 |
+
contour = contour[:, 0, :]
|
240 |
+
tck, _ = splprep([contour[:, 0], contour[:, 1]], s=smoothing_factor)
|
241 |
+
u = np.linspace(0, 1, num_points)
|
242 |
+
resampled_points = splev(u, tck)
|
243 |
+
smoothed_x = gaussian_filter1d(resampled_points[0], sigma=1)
|
244 |
+
smoothed_y = gaussian_filter1d(resampled_points[1], sigma=1)
|
245 |
+
return np.array([smoothed_x, smoothed_y]).T
|
246 |
+
|
247 |
+
# ---------------------
|
248 |
+
# Add the missing extract_outlines function
|
249 |
+
# ---------------------
|
250 |
+
def extract_outlines(binary_image: np.ndarray) -> (np.ndarray, list):
|
251 |
+
contours, _ = cv2.findContours(binary_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
|
252 |
+
outline_image = np.zeros_like(binary_image)
|
253 |
+
cv2.drawContours(outline_image, contours, -1, (255), thickness=2)
|
254 |
+
return cv2.bitwise_not(outline_image), contours
|
255 |
+
|
256 |
+
# ---------------------
|
257 |
+
# Functions for Finger Cut Clearance
|
258 |
+
# ---------------------
|
259 |
+
def union_tool_and_circle(tool_polygon: Polygon, center_inch, circle_diameter=1.0):
|
260 |
+
radius = circle_diameter / 2.0
|
261 |
+
circle_poly = Point(center_inch).buffer(radius, resolution=64)
|
262 |
+
union_poly = tool_polygon.union(circle_poly)
|
263 |
+
return union_poly
|
264 |
+
|
265 |
+
def build_tool_polygon(points_inch):
|
266 |
+
return Polygon(points_inch)
|
267 |
+
|
268 |
+
def polygon_to_exterior_coords(poly: Polygon):
|
269 |
+
if poly.geom_type == "MultiPolygon":
|
270 |
+
biggest = max(poly.geoms, key=lambda g: g.area)
|
271 |
+
poly = biggest
|
272 |
+
if not poly.exterior:
|
273 |
+
return []
|
274 |
+
return list(poly.exterior.coords)
|
275 |
+
|
276 |
+
def place_finger_cut_randomly(tool_polygon, points_inch, existing_centers, all_polygons, circle_diameter=1.0, min_gap=0.25, max_attempts=30):
|
277 |
+
import random
|
278 |
+
needed_center_distance = circle_diameter + min_gap
|
279 |
+
radius = circle_diameter / 2.0
|
280 |
+
for _ in range(max_attempts):
|
281 |
+
idx = random.randint(0, len(points_inch) - 1)
|
282 |
+
cx, cy = points_inch[idx]
|
283 |
+
too_close = False
|
284 |
+
for (ex_x, ex_y) in existing_centers:
|
285 |
+
if np.hypot(cx - ex_x, cy - ex_y) < needed_center_distance:
|
286 |
+
too_close = True
|
287 |
+
break
|
288 |
+
if too_close:
|
289 |
+
continue
|
290 |
+
circle_poly = Point((cx, cy)).buffer(radius, resolution=64)
|
291 |
+
union_poly = tool_polygon.union(circle_poly)
|
292 |
+
overlap_with_others = False
|
293 |
+
too_close_to_others = False
|
294 |
+
for poly in all_polygons:
|
295 |
+
if union_poly.intersects(poly):
|
296 |
+
overlap_with_others = True
|
297 |
+
break
|
298 |
+
if circle_poly.buffer(min_gap).intersects(poly):
|
299 |
+
too_close_to_others = True
|
300 |
+
break
|
301 |
+
if overlap_with_others or too_close_to_others:
|
302 |
+
continue
|
303 |
+
existing_centers.append((cx, cy))
|
304 |
+
return union_poly, (cx, cy)
|
305 |
+
print("Warning: Could not place a finger cut circle meeting all spacing requirements.")
|
306 |
+
return None, None
|
307 |
+
|
308 |
+
# ---------------------
|
309 |
+
# DXF Spline and Boundary Functions
|
310 |
+
# ---------------------
|
311 |
+
def save_dxf_spline(inflated_contours, scaling_factor, height, finger_clearance=False):
|
312 |
+
degree = 3
|
313 |
+
closed = True
|
314 |
+
doc = ezdxf.new(units=0)
|
315 |
+
doc.units = ezdxf.units.IN
|
316 |
+
doc.header["$INSUNITS"] = ezdxf.units.IN
|
317 |
+
msp = doc.modelspace()
|
318 |
+
finger_cut_centers = []
|
319 |
+
final_polygons_inch = []
|
320 |
+
for contour in inflated_contours:
|
321 |
+
try:
|
322 |
+
resampled_contour = resample_contour(contour)
|
323 |
+
points_inch = [(x * scaling_factor, (height - y) * scaling_factor) for x, y in resampled_contour]
|
324 |
+
if len(points_inch) < 3:
|
325 |
+
continue
|
326 |
+
if np.linalg.norm(np.array(points_inch[0]) - np.array(points_inch[-1])) > 1e-2:
|
327 |
+
points_inch.append(points_inch[0])
|
328 |
+
tool_polygon = build_tool_polygon(points_inch)
|
329 |
+
if finger_clearance:
|
330 |
+
union_poly, center = place_finger_cut_randomly(tool_polygon, points_inch, finger_cut_centers, final_polygons_inch, circle_diameter=1.0, min_gap=0.25, max_attempts=30)
|
331 |
+
if union_poly is not None:
|
332 |
+
tool_polygon = union_poly
|
333 |
+
exterior_coords = polygon_to_exterior_coords(tool_polygon)
|
334 |
+
if len(exterior_coords) < 3:
|
335 |
+
continue
|
336 |
+
msp.add_spline(exterior_coords, degree=degree, dxfattribs={"layer": "TOOLS"})
|
337 |
+
final_polygons_inch.append(tool_polygon)
|
338 |
+
except ValueError as e:
|
339 |
+
print(f"Skipping contour: {e}")
|
340 |
+
return doc, final_polygons_inch
|
341 |
+
|
342 |
+
def add_rectangular_boundary(doc, polygons_inch, boundary_length, boundary_width, boundary_unit):
|
343 |
+
msp = doc.modelspace()
|
344 |
+
if boundary_unit == "mm":
|
345 |
+
boundary_length_in = boundary_length / 25.4
|
346 |
+
boundary_width_in = boundary_width / 25.4
|
347 |
+
else:
|
348 |
+
boundary_length_in = boundary_length
|
349 |
+
boundary_width_in = boundary_width
|
350 |
+
min_x = float("inf")
|
351 |
+
min_y = float("inf")
|
352 |
+
max_x = -float("inf")
|
353 |
+
max_y = -float("inf")
|
354 |
+
for poly in polygons_inch:
|
355 |
+
b = poly.bounds
|
356 |
+
min_x = min(min_x, b[0])
|
357 |
+
min_y = min(min_y, b[1])
|
358 |
+
max_x = max(max_x, b[2])
|
359 |
+
max_y = max(max_y, b[3])
|
360 |
+
if min_x == float("inf"):
|
361 |
+
print("No tool polygons found, skipping boundary.")
|
362 |
+
return None
|
363 |
+
shape_cx = (min_x + max_x) / 2
|
364 |
+
shape_cy = (min_y + max_y) / 2
|
365 |
+
half_w = boundary_width_in / 2.0
|
366 |
+
half_l = boundary_length_in / 2.0
|
367 |
+
left = shape_cx - half_w
|
368 |
+
right = shape_cx + half_w
|
369 |
+
bottom = shape_cy - half_l
|
370 |
+
top = shape_cy + half_l
|
371 |
+
rect_coords = [(left, bottom), (right, bottom), (right, top), (left, top), (left, bottom)]
|
372 |
+
from shapely.geometry import Polygon as ShapelyPolygon
|
373 |
+
boundary_polygon = ShapelyPolygon(rect_coords)
|
374 |
+
msp.add_lwpolyline(rect_coords, close=True, dxfattribs={"layer": "BOUNDARY"})
|
375 |
+
return boundary_polygon
|
376 |
+
|
377 |
+
def draw_polygons_inch(polygons_inch, image_rgb, scaling_factor, image_height, color=(0,0,255), thickness=2):
|
378 |
+
for poly in polygons_inch:
|
379 |
+
if poly.geom_type == "MultiPolygon":
|
380 |
+
for subpoly in poly.geoms:
|
381 |
+
draw_single_polygon(subpoly, image_rgb, scaling_factor, image_height, color, thickness)
|
382 |
+
else:
|
383 |
+
draw_single_polygon(poly, image_rgb, scaling_factor, image_height, color, thickness)
|
384 |
+
|
385 |
+
def draw_single_polygon(poly, image_rgb, scaling_factor, image_height, color=(0,0,255), thickness=2):
|
386 |
+
ext = list(poly.exterior.coords)
|
387 |
+
if len(ext) < 3:
|
388 |
+
return
|
389 |
+
pts_px = []
|
390 |
+
for (x_in, y_in) in ext:
|
391 |
+
px = int(x_in / scaling_factor)
|
392 |
+
py = int(image_height - (y_in / scaling_factor))
|
393 |
+
pts_px.append([px, py])
|
394 |
+
pts_px = np.array(pts_px, dtype=np.int32)
|
395 |
+
cv2.polylines(image_rgb, [pts_px], isClosed=True, color=color, thickness=thickness, lineType=cv2.LINE_AA)
|
396 |
+
|
397 |
+
# ---------------------
|
398 |
+
# Main Predict Function with Finger Cut Clearance, Boundary Box, Annotation and Sharpness Enhancement
|
399 |
+
# ---------------------
|
400 |
+
def predict(
|
401 |
+
image: Union[str, bytes, np.ndarray],
|
402 |
+
offset_inches: float,
|
403 |
+
finger_clearance: str, # "Yes" or "No"
|
404 |
+
add_boundary: str, # "Yes" or "No"
|
405 |
+
boundary_length: float,
|
406 |
+
boundary_width: float,
|
407 |
+
boundary_unit: str,
|
408 |
+
annotation_text: str
|
409 |
+
):
|
410 |
+
overall_start = time.time()
|
411 |
+
# Convert image to NumPy array if needed.
|
412 |
+
if isinstance(image, str):
|
413 |
+
if os.path.exists(image):
|
414 |
+
image = np.array(Image.open(image).convert("RGB"))
|
415 |
+
else:
|
416 |
+
try:
|
417 |
+
image = np.array(Image.open(io.BytesIO(base64.b64decode(image))).convert("RGB"))
|
418 |
+
except Exception:
|
419 |
+
raise ValueError("Invalid base64 image data")
|
420 |
+
# Apply sharpness enhancement if image is a NumPy array.
|
421 |
+
if isinstance(image, np.ndarray):
|
422 |
+
pil_image = Image.fromarray(image)
|
423 |
+
enhanced_image = ImageEnhance.Sharpness(pil_image).enhance(1.5)
|
424 |
+
image = np.array(enhanced_image)
|
425 |
+
try:
|
426 |
+
t = time.time()
|
427 |
+
drawer_img = yolo_detect(image)
|
428 |
+
print("Drawer detection completed in {:.2f} seconds".format(time.time() - t))
|
429 |
+
t = time.time()
|
430 |
+
shrunked_img = make_square(shrink_bbox(drawer_img, 0.90))
|
431 |
+
del drawer_img
|
432 |
+
gc.collect()
|
433 |
+
print("Image shrinking completed in {:.2f} seconds".format(time.time() - t))
|
434 |
+
except DrawerNotDetectedError:
|
435 |
+
raise DrawerNotDetectedError("Drawer not detected! Please take another picture with a drawer.")
|
436 |
+
try:
|
437 |
+
t = time.time()
|
438 |
+
reference_obj_img, scaling_box_coords = detect_reference_square(shrunked_img)
|
439 |
+
print("Reference square detection completed in {:.2f} seconds".format(time.time() - t))
|
440 |
+
except ReferenceBoxNotDetectedError:
|
441 |
+
raise ReferenceBoxNotDetectedError("Reference box not detected! Please take another picture with a reference box.")
|
442 |
+
t = time.time()
|
443 |
+
reference_obj_img = make_square(reference_obj_img)
|
444 |
+
reference_square_mask = remove_bg_u2netp(reference_obj_img)
|
445 |
+
print("Reference image processing completed in {:.2f} seconds".format(time.time() - t))
|
446 |
+
t = time.time()
|
447 |
+
try:
|
448 |
+
cv2.imwrite("mask.jpg", cv2.cvtColor(reference_obj_img, cv2.COLOR_RGB2GRAY))
|
449 |
+
scaling_factor = calculate_scaling_factor(
|
450 |
+
reference_image_path="./Reference_ScalingBox.jpg",
|
451 |
+
target_image=reference_square_mask,
|
452 |
+
feature_detector="ORB",
|
453 |
+
)
|
454 |
+
except ZeroDivisionError:
|
455 |
+
scaling_factor = None
|
456 |
+
print("Error calculating scaling factor: Division by zero")
|
457 |
+
except Exception as e:
|
458 |
+
scaling_factor = None
|
459 |
+
print(f"Error calculating scaling factor: {e}")
|
460 |
+
if scaling_factor is None or scaling_factor == 0:
|
461 |
+
scaling_factor = 1.0
|
462 |
+
print("Using default scaling factor of 1.0 due to calculation error")
|
463 |
+
gc.collect()
|
464 |
+
print("Scaling factor determined: {}".format(scaling_factor))
|
465 |
+
t = time.time()
|
466 |
+
orig_size = shrunked_img.shape[:2]
|
467 |
+
objects_mask = remove_bg(shrunked_img)
|
468 |
+
processed_size = objects_mask.shape[:2]
|
469 |
+
objects_mask = exclude_scaling_box(objects_mask, scaling_box_coords, orig_size, processed_size, expansion_factor=1.2)
|
470 |
+
objects_mask = resize_img(objects_mask, (shrunked_img.shape[1], shrunked_img.shape[0]))
|
471 |
+
del scaling_box_coords
|
472 |
+
gc.collect()
|
473 |
+
print("Object masking completed in {:.2f} seconds".format(time.time() - t))
|
474 |
+
t = time.time()
|
475 |
+
offset_pixels = (offset_inches / scaling_factor) * 2 + 1 if scaling_factor != 0 else 1
|
476 |
+
dilated_mask = cv2.dilate(objects_mask, np.ones((int(offset_pixels), int(offset_pixels)), np.uint8))
|
477 |
+
del objects_mask
|
478 |
+
gc.collect()
|
479 |
+
print("Mask dilation completed in {:.2f} seconds".format(time.time() - t))
|
480 |
+
Image.fromarray(dilated_mask).save("./outputs/scaled_mask_new.jpg")
|
481 |
+
t = time.time()
|
482 |
+
outlines, contours = extract_outlines(dilated_mask)
|
483 |
+
shrunked_img_contours = cv2.drawContours(shrunked_img.copy(), contours, -1, (0, 0, 255), thickness=2)
|
484 |
+
del shrunked_img
|
485 |
+
gc.collect()
|
486 |
+
print("Outline extraction completed in {:.2f} seconds".format(time.time() - t))
|
487 |
+
t = time.time()
|
488 |
+
use_finger_clearance = True if finger_clearance.lower() == "yes" else False
|
489 |
+
doc, final_polygons_inch = save_dxf_spline(contours, scaling_factor, processed_size[0], finger_clearance=use_finger_clearance)
|
490 |
+
del contours
|
491 |
+
gc.collect()
|
492 |
+
print("DXF generation completed in {:.2f} seconds".format(time.time() - t))
|
493 |
+
boundary_polygon = None
|
494 |
+
if add_boundary.lower() == "yes":
|
495 |
+
boundary_polygon = add_rectangular_boundary(doc, final_polygons_inch, boundary_length, boundary_width, boundary_unit)
|
496 |
+
if boundary_polygon is not None:
|
497 |
+
final_polygons_inch.append(boundary_polygon)
|
498 |
+
# --- Annotation Text Placement (Bottom-Right) ---
|
499 |
+
min_x = float("inf")
|
500 |
+
min_y = float("inf")
|
501 |
+
max_x = -float("inf")
|
502 |
+
max_y = -float("inf")
|
503 |
+
for poly in final_polygons_inch:
|
504 |
+
b = poly.bounds
|
505 |
+
if b[0] < min_x:
|
506 |
+
min_x = b[0]
|
507 |
+
if b[1] < min_y:
|
508 |
+
min_y = b[1]
|
509 |
+
if b[2] > max_x:
|
510 |
+
max_x = b[2]
|
511 |
+
if b[3] > max_y:
|
512 |
+
max_y = b[3]
|
513 |
+
margin = 0.5
|
514 |
+
text_x = (min_x + max_x) / 2
|
515 |
+
text_y = min_y - margin
|
516 |
+
msp = doc.modelspace()
|
517 |
+
if annotation_text.strip():
|
518 |
+
text_entity = msp.add_text(
|
519 |
+
annotation_text.strip(),
|
520 |
+
dxfattribs={
|
521 |
+
"height": 0.25,
|
522 |
+
"layer": "ANNOTATION"
|
523 |
+
}
|
524 |
+
)
|
525 |
+
text_entity.dxf.insert = (text_x, text_y)
|
526 |
+
dxf_filepath = os.path.join("./outputs", "out.dxf")
|
527 |
+
doc.saveas(dxf_filepath)
|
528 |
+
# --- End Annotation Text Placement ---
|
529 |
+
draw_polygons_inch(final_polygons_inch, shrunked_img_contours, scaling_factor, processed_size[0], color=(0,0,255), thickness=2)
|
530 |
+
outlines_bgr = cv2.cvtColor(outlines, cv2.COLOR_GRAY2BGR)
|
531 |
+
draw_polygons_inch(final_polygons_inch, outlines_bgr, scaling_factor, processed_size[0], color=(0,0,255), thickness=2)
|
532 |
+
if annotation_text.strip():
|
533 |
+
text_px = int(text_x / scaling_factor)
|
534 |
+
text_py = int(processed_size[0] - (text_y / scaling_factor))
|
535 |
+
cv2.putText(shrunked_img_contours, annotation_text.strip(), (text_px, text_py), cv2.FONT_HERSHEY_SIMPLEX, 1, (0,0,255), 2, cv2.LINE_AA)
|
536 |
+
cv2.putText(outlines_bgr, annotation_text.strip(), (text_px, text_py), cv2.FONT_HERSHEY_SIMPLEX, 1, (0,0,255), 2, cv2.LINE_AA)
|
537 |
+
outlines_color = cv2.cvtColor(outlines_bgr, cv2.COLOR_BGR2RGB)
|
538 |
+
print("Total prediction time: {:.2f} seconds".format(time.time() - overall_start))
|
539 |
+
return (
|
540 |
+
cv2.cvtColor(shrunked_img_contours, cv2.COLOR_BGR2RGB),
|
541 |
+
outlines_color,
|
542 |
+
dxf_filepath,
|
543 |
+
dilated_mask,
|
544 |
+
str(scaling_factor)
|
545 |
+
)
|
546 |
+
|
547 |
+
# ---------------------
|
548 |
+
# Gradio Interface
|
549 |
+
# ---------------------
|
550 |
+
if __name__ == "__main__":
|
551 |
+
os.makedirs("./outputs", exist_ok=True)
|
552 |
+
def gradio_predict(img, offset, finger_clearance, add_boundary, boundary_length, boundary_width, boundary_unit, annotation_text):
|
553 |
+
return predict(img, offset, finger_clearance, add_boundary, boundary_length, boundary_width, boundary_unit, annotation_text)
|
554 |
+
iface = gr.Interface(
|
555 |
+
fn=gradio_predict,
|
556 |
+
inputs=[
|
557 |
+
gr.Image(label="Input Image"),
|
558 |
+
gr.Number(label="Offset value for Mask (inches)", value=0.075),
|
559 |
+
gr.Dropdown(label="Add Finger Clearance?", choices=["Yes", "No"], value="No"),
|
560 |
+
gr.Dropdown(label="Add Rectangular Boundary?", choices=["Yes", "No"], value="No"),
|
561 |
+
gr.Number(label="Boundary Length", value=300.0, precision=2),
|
562 |
+
gr.Number(label="Boundary Width", value=200.0, precision=2),
|
563 |
+
gr.Dropdown(label="Boundary Unit", choices=["mm", "inches"], value="mm"),
|
564 |
+
gr.Textbox(label="Annotation (max 20 chars)", max_length=20, placeholder="Type up to 20 characters")
|
565 |
+
],
|
566 |
+
outputs=[
|
567 |
+
gr.Image(label="Output Image"),
|
568 |
+
gr.Image(label="Outlines of Objects"),
|
569 |
+
gr.File(label="DXF file"),
|
570 |
+
gr.Image(label="Mask"),
|
571 |
+
gr.Textbox(label="Scaling Factor (inches/pixel)")
|
572 |
+
],
|
573 |
+
examples=[
|
574 |
+
["./examples/Test20.jpg", 0.075, "No", "No", 300.0, 200.0, "mm", "MyTool"],
|
575 |
+
["./examples/Test21.jpg", 0.075, "Yes", "Yes", 300.0, 200.0, "mm", "Tool2"]
|
576 |
+
]
|
577 |
+
)
|
578 |
+
iface.launch(share=True)
|