def submission(args):
logging.info('======= user config ======')
logging.info(pprint(opt))
logging.info(pprint(args))
logging.info('======= end ======')
test_loader, idx_to_class = get_test_provider(opt)
net = getattr(network, opt.network.name)(classes=opt.dataset.num_classes,
pretrain=False)
net.load_state_dict(torch.load(args.model_path)['state_dict'])
net.eval()
net = nn.DataParallel(net)
net = net.cuda()
preds = list()
img_ids = list()
for data, fname in test_loader:
with torch.no_grad():
scores = net(data)
pred_labels = scores.max(1)[1].cpu().numpy().tolist()
preds.extend(pred_labels)
img_ids.extend(fname)
df = pd.DataFrame({'file': img_ids, 'species': preds})
df['species'] = df['species'].apply(lambda x: idx_to_class[x])
df.to_csv('submission1.csv', index=False)
def submission(args):
test_loader, id_to_class = get_test_provider(args.bs)
net = network.ResNet18(num_classes=10)
net.load_state_dict(torch.load(args.model_path)['state_dict'])
net = nn.DataParallel(net)
net.eval()
if args.use_gpu:
net = net.cuda()
pred_labels = list()
indices = list()
for data, fname in test_loader:
if args.use_gpu:
data = data.cuda()
with torch.no_grad():
scores = net(data)
labels = scores.max(1)[1].cpu().numpy()
pred_labels.extend(labels)
indices.extend(fname.numpy())
df = pd.DataFrame({'id': indices, 'label': pred_labels})
df['label'] = df['label'].apply(lambda x: id_to_class[x])
df.to_csv('submission.csv', index=False)