def detect(checkpoint, input):
score_prediction = 0.5
chip_size = 300
#Histogram Equalize the image
img = cv2.imread(input, 0)
histImg = cv2.equalizeHist(img)
clahe3 = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))
cl13 = clahe3.apply(histImg)
#Parse and chip images
arr = np.array(np.expand_dims(cl13, axis=2))
chip_size = (chip_size,chip_size)
images = chip_image(arr,chip_size)
#generate detections
boxes, scores, classes = generate_detections(checkpoint,images)
#Process boxes to be full-sized
width,height,_ = arr.shape
cwn,chn = (chip_size)
wn,hn = (int(width/cwn),int(height/chn))
num_preds = 250
bfull = boxes[:wn*hn].reshape((wn,hn,num_preds,4))
b2 = np.zeros(bfull.shape)
b2[:,:,:,0] = bfull[:,:,:,1]
b2[:,:,:,1] = bfull[:,:,:,0]
b2[:,:,:,2] = bfull[:,:,:,3]
b2[:,:,:,3] = bfull[:,:,:,2]
bfull = b2
bfull[:,:,:,0] *= cwn
bfull[:,:,:,2] *= cwn
bfull[:,:,:,1] *= chn
bfull[:,:,:,3] *= chn
for i in range(wn):
for j in range(hn):
bfull[i,j,:,0] += j*cwn
bfull[i,j,:,2] += j*cwn
bfull[i,j,:,1] += i*chn
bfull[i,j,:,3] += i*chn
bfull = bfull.reshape((hn*wn,num_preds,4))
#get the file names
img_name = os.path.basename(input).split('.')[0]
model_name = os.path.basename(checkpoint).split('_')[1].split('.')[0]
file_name = img_name + '_' + model_name + '_preds.txt'
with open(file_name,'w') as f:
for i in range(bfull.shape[0]):
for j in range(bfull[i].shape[0]):
#box should be xmin ymin xmax ymax
box = bfull[i,j]
class_prediction = classes[i,j]
score_prediction = scores[i,j]
f.write('%d %d %d %d %d %f \n' % \
(box[0],box[1],box[2],box[3],int(class_prediction),score_prediction))
# Set up the dataframe from the txt file
data = pandas.read_csv(file_name, header=-1)
data.columns = ['init']
new = data["init"].str.split(" ", expand=True)
data["xmin"]= new[0].astype(int)
data["ymin"]= new[1].astype(int)
data["xmax"]= new[2].astype(int)
data["ymax"]= new[3].astype(int)
data["class_prediction"]= new[4].astype(int)
data["score_prediction"]= new[5].astype(float)
data.drop(["init"], axis=1, inplace = True)
# Confidence
data_con = data[data.score_prediction >= score_prediction]
# Set up the picture and draw the boxes
orgImg = cv2.imread(input)
# Save the image correctly
cv2.imwrite(img_name + '_' + model_name + '.png', orgImg)
name = img_name + '_' + model_name + '_' + 'preds.txt'
imagename = img_name + '_' + model_name + '.png'
return name, imagename
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("-c","--checkpoint", default='pbs/model.pb', help="Path to saved model")
parser.add_argument("-cs", "--chip_size", default=300, type=int, help="Size in pixels to chip input image")
parser.add_argument("input", help="Path to test chip")
parser.add_argument("-o","--output",default="predictions.txt",help="Filepath of desired output")
args = parser.parse_args()
#Parse and chip images
arr = np.array(Image.open(args.input))
chip_size = (args.chip_size,args.chip_size)
images = chip_image(arr,chip_size)
print(images.shape)
#generate detections
boxes, scores, classes = generate_detections(args.checkpoint,images)
#Process boxes to be full-sized
width,height,_ = arr.shape
cwn,chn = (chip_size)
wn,hn = (int(width/cwn),int(height/chn))
num_preds = 250
bfull = boxes[:wn*hn].reshape((wn,hn,num_preds,4))
b2 = np.zeros(bfull.shape)
b2[:,:,:,0] = bfull[:,:,:,1]
b2[:,:,:,1] = bfull[:,:,:,0]
b2[:,:,:,2] = bfull[:,:,:,3]
b2[:,:,:,3] = bfull[:,:,:,2]
bfull = b2
empty_image_idx = []
for idx, image in enumerate(im):
# skip chips that do not have buildings, avoid feeding them into inference
if len(box_chip[idx]) == 0 or (len(box_chip[idx]) == 1
and np.all(box_chip[idx] == 0)):
empty_image_idx.append(idx)
k = k + 1
else:
images_list.append(image)
images = np.array(images_list)
images.astype(np.uint8)
print('number of images without bboxes: ', k)
print('images shape: ', images.shape)
i = 0
boxes_pred, scores_pred, classes_pred = generate_detections(
args.checkpoint, images)
# debug
for idx, image in enumerate(im):
if idx in set(empty_image_idx):
box = np.zeros((1, num_preds, 4))
score = np.zeros((1, num_preds))
clss = np.zeros((1, num_preds))
boxes.append(box)
scores.append(score)
classes.append(clss)
else:
#continue
box_pred, score_pred, cls_pred = boxes_pred[i], scores_pred[
i], classes_pred[i]