def __init__(self, Net, epoch, batchSize, imgPath, tagPath, testTagPath,
testImgPath, testResult):
# multiprocessing.set_start_method('spawn', True)
'''
Net:PNet,RNet,ONet对应需要训练的网络名称
epoch,batchSize 批次和轮次
'''
self.writer = SummaryWriter(log_dir="./runs/{}_runs".format(Net))
self.netName = Net
self.testResult = testResult
self.device = deviceFun()
self.modelPath = str(
'/home/chinasilva/code/deeplearning_homework/project_5/model/' +
self.netName + '.pth')
if Net == 'PNet':
self.centlossSize = 32
self.size = 12
self.net = PNet().to(self.device)
elif Net == 'RNet':
self.centlossSize = 128
self.size = 24
self.net = RNet().to(self.device)
elif Net == 'ONet':
self.centlossSize = 256
self.size = 48
self.net = ONet().to(self.device)
else:
raise RuntimeError('训练时,请输入正确的网络名称')
self.batchSize = batchSize
self.epoch = epoch
self.myData = MyData(tagPath, imgPath)
self.testData = MyData(testTagPath, testImgPath)
self.lossFun1 = nn.MSELoss() #reduction='none'
self.lossFun2 = nn.MSELoss() #reduction='none'
self.lossFun3 = nn.MSELoss() #reduction='none'
# self.lossFun1=nn.BCELoss()
# self.lossFun2=nn.MSELoss()
# self.lossFun3=nn.MSELoss()
self.criterion = nn.CrossEntropyLoss()
self.center_loss_layer = CenterLoss(2, self.centlossSize)
if os.path.exists(self.modelPath):
self.net = torch.load(self.modelPath)
# self.optimizer=torch.optim.SGD(self.net.parameters(), lr=0.001,momentum=0.9,weight_decay=0.00003)
# self.optimizer3 = torch.optim.SGD(self.net.parameters(),lr=0.5)
# self.optimizer3 = torch.optim.Adam(self.center_loss_layer.parameters())
# self.optimizer=torch.optim.Adam(self.net.parameters())
self.optimizer = torch.optim.SGD(self.net.parameters(),
lr=0.0001,
momentum=0.9,
weight_decay=0.0005)
self.optimizer2 = Lookahead(self.optimizer, k=5,
alpha=0.5) # Initialize Lookahead
def __init__(self, batchSize, epoch):
self.device = deviceFun()
self.batchSize = batchSize
self.epoch = epoch
self.myData = MyData()
self.preModelLoction = PRE_MODEL_PATH
self.lossFun = nn.MSELoss()
self.net = MyNet(cls_num=CLASS_NUM).to(self.device)
self.darkNet = DarkNet().to(self.device)
self.loadModel(self.preModelLoction)
self.optimizer = torch.optim.Adam(
filter(lambda p: p.requires_grad != None, self.net.parameters()))
def __init__(self, testImagePath, netPath):
self.device = deviceFun(cpu=False)
self.pnet = torch.load(netPath + '/PNet.pth') #,map_location="cpu"
self.rnet = torch.load(netPath + '/RNet.pth')
self.onet = torch.load(netPath + '/ONet.pth')
self.testImagePath = testImagePath
self.pnetSize = 12
self.rnetSize = 24
self.onetSize = 48
self.pnet.eval()
self.rnet.eval()
self.onet.eval()
self.imgName = ''
def __init__(self):
super().__init__()
self.device = deviceFun()
self.net = MyNet()
# self.loadModel()
self.net = self.net.load_state_dict(
torch.load(PRE_MODEL_PATH, map_location=self.device))
# torch.load(PRE_MODEL_PATH,)
self.trainImagePath = IMG_PATH
self.net.eval()
self.imgName = ''
self.net = torch.load(MODEL_PATH)
self.trainImagePath = IMG_PATH
self.net.eval()
self.imgName = ''
def __init__(self,
in_features,
out_features,
s=2.0,
m=0.50,
easy_margin=False): #m=0.50
super(ArcMarginProduct, self).__init__()
self.device = deviceFun()
self.in_features = in_features
self.out_features = out_features
self.s = s
self.m = m
self.weight = nn.Parameter(torch.FloatTensor(out_features,
in_features))
nn.init.xavier_uniform_(self.weight)
self.easy_margin = easy_margin
self.cos_m = math.cos(m)
self.sin_m = math.sin(m)
self.th = math.cos(math.pi - m)
self.mm = math.sin(math.pi - m) * m
def __init__(self, ratio):
super(NLL_OHEM, self).__init__(None, True)
self.ratio = ratio
self.device = deviceFun()