nathbns's picture
Upload laps.py
e9341b1 verified
import deps
import numpy as np
import cv2
import collections
import scipy
import scipy.cluster
import tensorflow as tf
import os
model_h5_path = "data/laps_models/laps.h5"
try:
NEURAL_MODEL = tf.keras.models.load_model(model_h5_path, compile=False)
from tensorflow.keras.optimizers import RMSprop
NEURAL_MODEL.compile(RMSprop(learning_rate=0.001),
loss='categorical_crossentropy',
metrics=['categorical_accuracy'])
except Exception as e:
print(f"Warning: Could not load model from {model_h5_path}: {e}")
from deps.laps import model as NEURAL_MODEL
def laps_intersections(lines):
__lines = [[(a[0], a[1]), (b[0], b[1])] for a, b in lines]
return deps.geometry.isect_segments(__lines)
def laps_cluster(points, max_dist=10):
Y = scipy.spatial.distance.pdist(points)
Z = scipy.cluster.hierarchy.single(Y)
T = scipy.cluster.hierarchy.fcluster(Z, max_dist, 'distance')
clusters = collections.defaultdict(list)
for i in range(len(T)):
clusters[T[i]].append(points[i])
clusters = clusters.values()
clusters = map(lambda arr: (np.mean(np.array(arr)[:, 0]),
np.mean(np.array(arr)[:, 1])), clusters)
return list(clusters)
def laps_detector(img):
global NC_LAYER
hashid = str(hash(img.tostring()))
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
img = cv2.threshold(img, 0, 255, cv2.THRESH_OTSU)[1]
img = cv2.Canny(img, 0, 255)
img = cv2.resize(img, (21, 21), interpolation=cv2.INTER_CUBIC)
imgd = img
X = [np.where(img > int(255/2), 1, 0).ravel()]
X = X[0].reshape([-1, 21, 21, 1])
img = cv2.dilate(img, None)
mask = cv2.copyMakeBorder(img, top=1, bottom=1, left=1, right=1,
borderType=cv2.BORDER_CONSTANT, value=[255, 255, 255])
mask = cv2.bitwise_not(mask)
i = 0
contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_NONE)
_c = np.zeros((23, 23, 3), np.uint8)
for cnt in contours:
(x, y), radius = cv2.minEnclosingCircle(cnt)
x, y = int(x), int(y)
approx = cv2.approxPolyDP(cnt, 0.1*cv2.arcLength(cnt, True), True)
if len(approx) == 4 and radius < 14:
cv2.drawContours(_c, [cnt], 0, (0, 255, 0), 1)
i += 1
else:
cv2.drawContours(_c, [cnt], 0, (0, 0, 255), 1)
if i == 4:
return (True, 1)
pred = NEURAL_MODEL.predict(X)
a, b = pred[0][0], pred[0][1]
t = a > b and b < 0.03 and a > 0.975
if t:
return (True, pred[0])
else:
return (False, pred[0])
################################################################################
def LAPS(img, lines, size=10):
__points, points = laps_intersections(lines), []
for pt in __points:
pt = list(map(int, pt))
lx1 = max(0, int(pt[0]-size-1))
lx2 = max(0, int(pt[0]+size))
ly1 = max(0, int(pt[1]-size))
ly2 = max(0, int(pt[1]+size+1))
dimg = img[ly1:ly2, lx1:lx2]
dimg_shape = np.shape(dimg)
if dimg_shape[0] <= 0 or dimg_shape[1] <= 0:
continue
re_laps = laps_detector(dimg)
if not re_laps[0]:
continue
if pt[0] < 0 or pt[1] < 0:
continue
points += [pt]
points = laps_cluster(points)
return points