# full_name = "/home/yuliang/code/MobilePose-pytorch/models/demo/resnet18_227x227-robust.t7" # Rescale Expansion ToTensor
# full_name = "/home/yuliang/code/MobilePose-pytorch/models/demo/resnet18_227x227.t7" # Rescale Expansion ToTensor
ROOT_DIR = "../deeppose_tf/datasets/mpii"
if modeltype == 'resnet':
full_name = "/home/yuliang/code/MobilePose-pytorch/models/demo/resnet18_227x227.t7" # Rescale Expansion ToTensor
input_size = 227
# test_dataset = PoseDataset(csv_file=os.path.join(ROOT_DIR,'test_joints.csv'),
# transform=transforms.Compose([
# Rescale((input_size, input_size)), # resnet use
# # Wrap((input_size,input_size)), # mobilenet use
# Expansion(),
# ToTensor()
# ]))
test_dataset = DatasetFactory.get_test_dataset(modeltype, input_size)
elif modeltype == 'mobilenet':
full_name = "/home/yuliang/code/MobilePose-pytorch/models/demo/mobilenetv2_224x224-robust.t7" # Wrap Expansion ToTensor
input_size = 224
# test_dataset = PoseDataset(csv_file=os.path.join(ROOT_DIR,'test_joints.csv'),
# transform=transforms.Compose([
# Rescale((input_size, input_size)), # resnet use
# # Wrap((input_size,input_size)), # mobilenet use
# Expansion(),
# ToTensor()
# ]))
test_dataset = DatasetFactory.get_test_dataset(modeltype, input_size)
print("Loading testing dataset, wait...")
parser.add_argument('--t7', type=str, required=True, default="")
parser.add_argument('--gpu', type=str, required=True, default="")
args = parser.parse_args()
modelpath = args.t7
device = torch.device("cuda" if len(args.gpu) > 0 else "cpu")
# user defined parameters
num_threads = multiprocessing.cpu_count()
PATH_PREFIX = "./results/{}".format(modelpath.split(".")[0])
input_size = 224
modelname = args.model
test_dataset = DatasetFactory.get_test_dataset("resnet", input_size)
print("Loading testing dataset, wait...")
bs_test = len(test_dataset)
test_dataloader = DataLoader(test_dataset,
batch_size=bs_test,
shuffle=False,
num_workers=num_threads)
# get all test data
all_test_data = {}
for i_batch, sample_batched in enumerate(tqdm(test_dataloader)):
all_test_data = sample_batched
eval_coco(all_test_data, modelname, modelpath,
os.path.join(PATH_PREFIX, 'result-gt-json.txt'),
os.path.join(PATH_PREFIX, 'result-pred-json.txt'))
# net = torch.load('./models/%s/%s'%(modeltype,modelname)).cuda()
net = torch.load('./models/%s/%s' % (modeltype, modelname)).cuda()
# alog.info(net)
net = net.train()
ROOT_DIR = "../deeppose_tf/datasets/mpii" # root dir to the dataset
PATH_PREFIX = './models/{}/'.format(modeltype) # path to save the model
tmp_modeltype = "resnet"
train_dataset = DatasetFactory.get_train_dataset(tmp_modeltype, inputsize)
train_dataloader = DataLoader(train_dataset,
batch_size=batchsize,
shuffle=False,
num_workers=num_threads)
test_dataset = DatasetFactory.get_test_dataset(tmp_modeltype, inputsize)
test_dataloader = DataLoader(test_dataset,
batch_size=batchsize,
shuffle=False,
num_workers=num_threads)
criterion = nn.MSELoss().cuda()
# optimizer = optim.Adam(net.parameters(), lr=learning_rate, betas=(0.9, 0.999), eps=1e-08)
# optimizer = optim.SGD(net.parameters(), lr=learning_rate, momentum=0.9)
optimizer = optim.RMSprop(net.parameters(), lr=learning_rate, momentum=0.9)
def mse_loss(input, target):
return torch.sum(torch.pow(input - target, 2)) / input.nelement()
train_loss_all = []
valid_loss_all = []