model = nn.Sequential()
model.add_module('conv_1', nn.Conv2d(in_channels=3, out_channels=16, kernel_size=3))
model.add_module('bn_1',nn.BatchNorm2d(16))
model.add_module('relu_1', nn.ReLU())
model.add_module('conv_1_1', nn.Conv2d(in_channels=16, out_channels=64, kernel_size=3))
model.add_module('bn_1_1', nn.BatchNorm2d(64))
model.add_module('relu_1', nn.ReLU())
model.add_module('mp_1', nn.MaxPool2d(2))
model.add_module('conv_2', nn.Conv2d(in_channels=64, out_channels=128, kernel_size=3))
model.add_module('bn_2', nn.BatchNorm2d(128))
model.add_module('relu_2', nn.ReLU())
model.add_module('mp_2', nn.MaxPool2d(2))
model.add_module('conv_3', nn.Conv2d(in_channels=128, out_channels=256, kernel_size=3))
model.add_module('bn_3', nn.BatchNorm2d(256))
model.add_module('relu_3', nn.ReLU())
model.add_module('mp_3', nn.MaxPool2d(2))
model.add_module('flat', Flatten())
model.add_module('dropout_1', nn.Dropout(0.25))
model.add_module('linear_1', nn.Linear(256 * 6 * 6, 8000))
model.add_module('bn_4', nn.BatchNorm1d(8000))
model.add_module('relu_4', nn.ReLU())
model.add_module('linear_2', nn.Linear(8000, 4000))
model.add_module('relu_5', nn.ReLU())
model.add_module('linear_3', nn.Linear(4000, 2000))
model.add_module('bn_5', nn.BatchNorm1d(2000))
model.add_module('relu_6', nn.ReLU())
model.add_module('dropout_1', nn.Dropout(0.25))
model.add_module('dense4', nn.Linear(2000, 1000))
model.add_module('bn_6', nn.BatchNorm1d(1000))
model.add_module('relu_7', nn.ReLU())
model.add_module('dropout_2', nn.Dropout(0.25))
model.add_module('relu_8', nn.ReLU())
model.add_module('logits', nn.Linear(1000, 200))
model.load_state_dict(torch.load('best_model.sav'))
model.eval()
model.cuda()