def __init__(self, channels):
super(YOLO_PRECONNECT, self).__init__()
self.conv1 = utils.Conv2d(channels,
64, (3, 3),
padding=1,
pool=True,
pool_size=(2, 4))
self.conv2 = utils.Conv2d(64,
128, (3, 3),
padding=1,
pool=True,
pool_size=(2, 4))
self.conv3 = utils.Conv2d(128, 256, (3, 3), padding=1, pool=False)
#self.conv4 = utils.Conv2d(256,512,(3,3),padding=1,pool=True,pool_size=(2,4))
#self.conv5 = utils.Conv2d(512,256,(1,1),pool=False)
#self.conv6 = utils.Conv2d(256,512,(3,3),padding=1,pool=False)
#self.conv7 = utils.Conv2d(512,256,(1,1),pool=False)
self.conv8 = utils.Conv2d(256,
512, (3, 3),
padding=1,
pool=True,
pool_size=(2, 4))
self.output_size = 512
self.reduction_factor = (2 * 2 * 2, 4 * 4 * 4)
def __init__(self, nChannels, nPoints, nCoords, nClasses):
super(PointNet2d, self).__init__()
self.input_trans = Transform2d(nPoints,
nCoords,
nChannels,
3,
initial_reshape=True)
self.conv64A = utils.Conv2d(nChannels, 64, (1, nCoords), pool=False)
self.conv64B = utils.Conv2d(64, 64, (1, 1), pool=False)
self.feature_trans = Transform2d(nPoints, 1, 64, 64)
self.conv64C = utils.Conv2d(64, 64, (1, 1), pool=False)
self.conv128 = utils.Conv2d(64, 128, (1, 1), pool=False)
self.conv1024 = utils.Conv2d(128, 1024, (1, 1), pool=False)
self.pool = torch.nn.MaxPool2d((nPoints, 1))
self.linear512 = utils.Linear(1024, 512)
self.dropoutA = torch.nn.Dropout(0.7)
self.linear256 = utils.Linear(512, 256)
self.dropoutB = torch.nn.Dropout(0.7)
self.linearID = utils.Linear(256, nClasses, bn=False, activation=None)
def __init__(self, channels):
super(YOLO_POSTCONNECT, self).__init__()
self.conv1 = utils.Conv2d(channels,
1024, (3, 3),
padding=1,
pool=False)
self.conv2 = utils.Conv2d(1024, 512, (1, 1), pool=False)
self.conv3 = utils.Conv2d(512, 1024, (3, 3), padding=1, pool=False)
self.conv4 = utils.Conv2d(1024, 512, (1, 1), pool=False)
self.conv5 = utils.Conv2d(512, 1024, (3, 3), padding=1, pool=False)
self.output_size = 1024
self.reduction_factor = (1, 1)
def __init__(self, height, width, channels, K, initial_reshape=False):
super(Transform2d, self).__init__()
self.K = K
self.initial_reshape = initial_reshape
self.conv64 = utils.Conv2d(channels, 64, (1, width), pool=False)
self.conv128 = utils.Conv2d(64, 128, (1, 1), pool=False)
self.conv1024 = utils.Conv2d(128, 1024, (1, 1), pool=False)
self.pool = torch.nn.MaxPool2d((height, 1))
self.linear512 = utils.Linear(1024, 512)
self.linear256 = utils.Linear(512, 256)
self.weights = torch.zeros(256, K * K, requires_grad=True)
self.biases = torch.eye(K, requires_grad=True).flatten()
def __init__(self, config):
super(YOLOClassOnly, self).__init__()
img_shape = config['data_handling']['image_shape']
num_classes = len(config['data_handling']['classes'])
self.pre_connect = YOLO_PRECONNECT(img_shape[0])
self.post_connect = YOLO_POSTCONNECT(self.pre_connect.output_size)
self.conv_connect = utils.Conv2d(self.pre_connect.output_size +
self.post_connect.output_size,
1024, (3, 3),
padding=1,
pool=False)
self.feature_layer = utils.Conv2d(1024,
1 + num_classes, (3, 3),
padding=1,
pool=False)
reduction_factor = (self.pre_connect.reduction_factor[0] *
self.post_connect.reduction_factor[0],
self.pre_connect.reduction_factor[1] *
self.post_connect.reduction_factor[1])
gridh = img_shape[1] // reduction_factor[0]
gridw = img_shape[2] // reduction_factor[1]
self.output_grid = (gridh, gridw)
logger.info('grid size = %s x %s', self.output_grid[0],
self.output_grid[1])
if YOLOClassOnly.BCEWithLogitsLoss is None:
pos_weight = torch.Tensor(gridh, gridw).fill_(gridh * gridw)
YOLOClassOnly.BCEWithLogitsLoss = torch.nn.BCEWithLogitsLoss(
pos_weight=pos_weight)