def create_dummy_predictions()

in mapillary_sls/utils/eval.py [0:0]

• 24 lines of code
• 10 McCabe index (conditional complexity)

def create_dummy_predictions(prediction_path, dataset):
    print("==> Prediction file {} doesn't exist".format(prediction_path))
    print("==> We create a new dummy prediction file at :")
    print("==> ", prediction_path)

    numQ = len(dataset.qIdx)
    numDb = len(dataset.dbImages)

    # all keys in the database and query keys
    query_keys = np.asarray([','.join([bn(k)[:-4] for k in key.split(',')]) for key in dataset.qImages[dataset.qIdx]]).reshape(numQ,1)
    database_keys = [','.join([bn(k)[:-4] for k in key.split(',')]) for key in dataset.dbImages]
    # choose n = min(5, numDb) random elements from the database
    ranks = np.asarray([np.random.choice(database_keys, replace=False, size = (min(5, numDb))) for q in range(numQ)])

    if ',' in query_keys[0,0]:
        qtxt = f"sequence with keys {query_keys[0,0]}"
    else:
        qtxt = f"image with key {query_keys[0, 0]}"
    mtxt = ' '.join([str(i) for i in ranks[0][:-1]]) + ' and ' + str(ranks[0][-1]) + '.'
    if ',' in ranks[0,0]:
        mtxt = f"{len(ranks[0])} sequences: " + mtxt
    else:
        mtxt = f"{len(ranks[0])} images: " + mtxt
    hdr = ("Each row contains the key of a query image or a comma-separated set of keys for a query sequence,\n"
           "followed by space-separated predicted image keys or sets of comma-separated keys for predicted sequences\n"
           "The format is valid for all tasks (im2im, im2seq, seq2im, seq2seq)\n"
           "For example, in this file, the " + qtxt + " is predicted to match with " + mtxt)

    # save the dummy predictions
    np.savetxt(prediction_path, np.concatenate([query_keys, ranks], axis=1), fmt='%s',
               header=hdr)