def load_config(config_path):
trn_opt_data = json.load(open(opt_path))
trn_opt = OptConfig()
trn_opt.load(trn_opt_data)
load_dim(trn_opt)
trn_opt.gpu_ids = opt.gpu_ids
trn_opt.dataroot = 'dataset/wild'
trn_opt.serial_batches = True
if not hasattr(trn_opt,
'normalize'): # previous model has no attribute normalize
setattr(trn_opt, 'normalize', False)
if not hasattr(trn_opt, 'loss_type'):
setattr(trn_opt, 'loss_type', 'mse')
return trn_opt
def load_from_opt_record(file_path):
opt_content = json.load(open(file_path, 'r'))
opt = OptConfig()
opt.load(opt_content)
return opt
if __name__ == '__main__':
# opt = TrainOptions().parse() # get training options
# opt.isTrain = False # set isTrain = False
model_name = 'ef_AVL_Adnn512,256,128_Vlstm128_maxpool_Lcnn128_fusion256,128run{}'
total_val = []
total_tst = []
for cv in range(1, 11):
models = []
for run_idx in range(1, 5):
cur_model_name = model_name.format(run_idx)
opt_info = json.load(
open(
'checkpoints/ef_AVL_Adnn512,256,128_Vlstm128_maxpool_Lcnn128_fusion256,128run1/train_opt.conf'
))
opt = OptConfig()
opt.load(opt_info)
opt.isTrain = False
opt.gpu_ids = [0]
opt.cvNo = cv
model = create_model(
opt) # create a model given opt.model and other options
model.setup(
opt
) # regular setup: load and print networks; create schedulers
model.load_networks_cv('checkpoints/{}/{}'.format(
cur_model_name, cv))
model.cuda()
model.eval()
models.append(model)