Main Result changing bugfix

[FredericOdermatt]

    print(np.unique(nmi_gts), np.unique(nmi_preds), np.unique(nmi_preds_w_bg))
    if is_full:
        nmi = normalized_mutual_info_score(nmi_gts, nmi_preds_w_bg) * 100
        ari = adjusted_rand_score(nmi_gts, nmi_preds_w_bg) * 100
    else:
        nmi = normalized_mutual_info_score(nmi_gts, nmi_preds) * 100
        ari = adjusted_rand_score(nmi_gts, nmi_preds_w_bg) * 100
    return nmi, ari

ari = adjusted_rand_score(nmi_gts, nmi_preds_w_bg) * 100 --> ari = adjusted_rand_score(nmi_gts, nmi_preds) * 100

in the evaluation notebook the foreground ari score is calculated with nmi predictions with background. FG-ARI on CUB increases from 21.0 to 21.65 when fixed. As this is a main result I think it is important to change this in the code and paper

[subhc]

Thank you! Really appreciate this fix.