ucohen

it seems like the loss function does not match this dataset, as there are numerous items with no attributes - causing zero effect on the loss. Any suggestions?

got the same issue, it seems like there is an offset to count for matlab's outputs. see: https://github.com/cheind/py-motmetrics/blob/333495ea95b33ab0218885a2e95cbcfd86a4399e/motmetrics/io.py#L87 it would be best to get an options flag for it.