evaluation code

[hoyeYang]

Amazing work! Could you provide the code for evaluating the model?

[evonneng]

Hi! Thank you for your interest! And yes, this is something I can add to the repo as a todo. Thanks for flagging this!

But to help in unblocking you for the time being, to implement the FD score that is in the paper, we adapt this function but instead of calculating the activation statistics, we calculate the statistics from the raw poses.

https://github.com/hukkelas/pytorch-frechet-inception-distance/blob/master/fid.py#L108

Diversity for the static poses was computed with 1) just taking the variance across a temporal sequence and 2) using this function:

def calculate_diversity(activation, diversity_times=10_000):
    assert len(activation.shape) == 2
    assert activation.shape[0] > diversity_times
    num_samples = activation.shape[0]
    first_indices = np.random.choice(num_samples, diversity_times, replace=False)
    second_indices = np.random.choice(num_samples, diversity_times, replace=False)
    dist = linalg.norm(activation[first_indices] - activation[second_indices], axis=1)
    return dist

Hope this helps in the meantime! Please let me know if there is any questions in the meantime.

[liujf69]

Thank you for your outstanding work. I am new in this field. Could you please provide all the codes for calculating evaluation indicators?

[evonneng]

Hi! thank you for your interest in this work! Providing the full evaluation indicators will be on our list of todo's. I just have to clean it up a bit, but will push a PR soon hopefully!