eval_retrieve_knn_pred.py [46:67]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    if train_padding_num > 0:
        for i in range(train_padding_num, args.num_replica):
            feat_train[i] = feat_train[i][:-1, :]
            feat_train_cls[i] = feat_train_cls[i][:-1]
    feat_train = np.concatenate(feat_train, axis=0).squeeze()
    feat_train_cls = np.concatenate(feat_train_cls, axis=0).squeeze()
    print('feat_train: {}'.format(feat_train.shape))
    print('feat_train_cls: {}'.format(feat_train_cls.shape))

    feat_val = []
    feat_val_cls = []
    for i in range(args.num_replica):
        feat_val.append(np.load(os.path.join(args.output_dir, 'feature_{}_{}.npy'.format(args.valsplit, i))))
        feat_val_cls.append(np.load(os.path.join(args.output_dir, 'feature_{}_cls_{}.npy'.format(args.valsplit, i))))
    if val_padding_num > 0:
        for i in range(val_padding_num, args.num_replica):
            feat_val[i] = feat_val[i][:-1, :]
            feat_val_cls[i] = feat_val_cls[i][:-1]
    feat_val = np.concatenate(feat_val, axis=0)
    feat_val_cls = np.concatenate(feat_val_cls, axis=0)
    print('feat_val: {}'.format(feat_val.shape))
    print('feat_val_cls: {}'.format(feat_val_cls.shape))
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



eval_svm_feature_perf.py [43:64]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    if train_padding_num > 0:
        for i in range(train_padding_num, args.num_replica):
            feat_train[i] = feat_train[i][:-1, :]
            feat_train_cls[i] = feat_train_cls[i][:-1]
    feat_train = np.concatenate(feat_train, axis=0).squeeze()
    feat_train_cls = np.concatenate(feat_train_cls, axis=0).squeeze()
    print('feat_train: {}'.format(feat_train.shape))
    print('feat_train_cls: {}'.format(feat_train_cls.shape))

    feat_val = []
    feat_val_cls = []
    for i in range(args.num_replica):
        feat_val.append(np.load(os.path.join(args.output_dir, 'feature_{}_{}.npy'.format(args.valsplit, i))))
        feat_val_cls.append(np.load(os.path.join(args.output_dir, 'feature_{}_cls_{}.npy'.format(args.valsplit, i))))
    if val_padding_num > 0:
        for i in range(val_padding_num, args.num_replica):
            feat_val[i] = feat_val[i][:-1, :]
            feat_val_cls[i] = feat_val_cls[i][:-1]
    feat_val = np.concatenate(feat_val, axis=0)
    feat_val_cls = np.concatenate(feat_val_cls, axis=0)
    print('feat_val: {}'.format(feat_val.shape))
    print('feat_val_cls: {}'.format(feat_val_cls.shape))
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -