def main():
#get user options/configurations
opt = opts.get_opt()
#set cuda visible devices if user specified a value
if opt.gpus is not None:
os.environ['CUDA_VISIBLE_DEVICES'] = opt.gpus
#load class to save metrics, iamges and models to disk
output = outputs.Outputs(opt)
#load class to store metrics and losses values
metric = metrics.Metrics()
#load synthetic dataset
healthy_dataloader_train, anomaly_dataloader_train = init_synth_dataloader_original(
opt.folder_toy_dataset, opt.batch_size, mode='train')
healthy_dataloader_val, anomaly_dataloader_val = init_synth_dataloader_original(
opt.folder_toy_dataset, opt.batch_size, mode='val')
net_g, net_r = init_model(opt)
optim_g, optim_r = init_optimizer(opt, net_g=net_g, net_r=net_r)
net_g = net_g.cuda()
net_r = net_r.cuda()
train(opt,
healthy_dataloader_train,
anomaly_dataloader_train,
healthy_dataloader_val,
anomaly_dataloader_val,
net_g=net_g,
net_r=net_r,
optim_g=optim_g,
optim_r=optim_r,
output=output,
metric=metric)
def main():
"""Train model and run inference on coco test set to output metrics"""
args = get_opt()
os.environ["CUDA_VISIBLE_DEVICES"] = str(args.gpu)
if(args.is_train == 1):
if not os.path.exists(args.model_dir):
os.makedirs(args.model_dir)
train(args)
bestmodelfn = os.path.join(args.model_dir, 'model.pth')
if(os.path.exists(bestmodelfn)):
if(args.beam_size == 1):
scores = test(args, 'test', modelfn=bestmodelfn)
else:
scores = test_beam(args, 'test', modelfn=bestmodelfn)
print('TEST set scores')
for k, v in list(scores[0].items()):
print(('%s: %f' % (k, v)))
else:
raise Exception('No checkpoint found %s' % bestmodelfn)
if args.caption_model == "transformer":
optim_state_dict = optimizer.optimizer.state_dict()
else:
optim_state_dict = optimizer.state_dict()
torch.save(
{
'epoch': epoch,
'decoder_state_dict': decoder.state_dict(),
'encoder_state_dict': encoder.state_dict(),
'optimizer': optim_state_dict,
'iteration': iteration,
args.score_select: score,
}, save_path)
if (score > bestscore):
bestscore = score
print(('[DEBUG] Saving model at epoch %d with %s score of %f'\
% (epoch, args.score_select, score)))
bestmodel_path = os.path.join(args.model_path, 'best_model.ckpt')
os.system('cp %s %s' % (save_path, bestmodel_path))
# test
args.vis = True
test(args, "test", encoder=encoder, decoder=decoder)
if __name__ == '__main__':
args = get_opt()
print(args)
main(args)