if __name__ == "__main__":
import random
from core import CenterNet, DetectionDataset, CenterValidTransform
from core import Prediction
from collections import OrderedDict
input_size = (512, 512)
root = os.path.dirname(
os.path.dirname(
os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
transform = CenterValidTransform(input_size,
mean=(0.485, 0.456, 0.406),
std=(0.229, 0.224, 0.225),
make_target=False)
dataset = DetectionDataset(path=os.path.join(root, 'valid'),
transform=transform)
num_classes = dataset.num_class
name_classes = dataset.classes
length = len(dataset)
image, label, _, _, _ = dataset[random.randint(0, length - 1)]
net = CenterNet(base=18,
heads=OrderedDict([('heatmap', {
'num_output': num_classes,
'bias': -2.19
}), ('offset', {
'num_output': 2
}), ('wh', {
'num_output': 2
})]),
head_conv_channel=64,
return ids, scores, bboxes
# test
if __name__ == "__main__":
from core import Efficient, EfficientTrainTransform, DetectionDataset
import os
input_size = (512, 512)
root = os.path.dirname(
os.path.dirname(
os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
transform = EfficientTrainTransform(input_size[0],
input_size[1],
make_target=False)
dataset = DetectionDataset(path=os.path.join(root, 'Dataset', 'train'),
transform=transform)
num_classes = dataset.num_class
image, label, _, _, _ = dataset[0]
label = mx.nd.array(label)
net = Efficient(
version=0,
input_size=input_size,
anchor_sizes=[32, 64, 128, 256, 512],
anchor_size_ratios=[1, pow(2, 1 / 3), pow(2, 2 / 3)],
anchor_aspect_ratios=[0.5, 1, 2],
num_classes=num_classes, # foreground만
anchor_box_offset=(0.5, 0.5),
anchor_box_clip=True,
alloc_size=[256, 256],
ctx=mx.cpu())
index = np.where(recall[1:] != recall[:-1])[0]
AP = np.sum((recall[index + 1] - recall[index]) * precision[index + 1])
return name, AP
if __name__ == "__main__":
import random
from core import SSD_VGG16, DetectionDataset
from core import Prediction
input_size = (512, 512)
root = os.path.dirname(
os.path.dirname(
os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
dataset = DetectionDataset(path=os.path.join(root, 'Dataset', 'train'),
input_size=input_size,
box_normalization=True)
num_classes = dataset.num_class
name_classes = dataset.classes
length = len(dataset)
image, label, _ = dataset[random.randint(0, length - 1)]
net = SSD_VGG16(
height_width_size=input_size,
box_sizes=[21, 51.2, 133.12, 215.04, 296.96, 378.88, 460.8, 542.72],
box_ratios=[[1, 2, 0.5]] + # conv4_3
[[1, 2, 0.5, 3, 1.0 / 3]] * 4 + # conv7, conv8_2, conv9_2, conv10_2
[[1, 2, 0.5]] * 2, # conv11_2, conv12_2
num_classes=num_classes,
# feature_sizes=[38, 19, 10, 5, 3, 1], # input_size : 300 - 6개
feature_sizes=[64, 32, 16, 8, 4, 2, 1], # input_size : 512 - 7개
