AFD#
Classes#
- class models.twf_utils.afd.ChannelAttn(c, n_tasks, reduction_rate=1, activated_with_softmax=False)[source]#
Bases:
Module
- class models.twf_utils.afd.ConditionalLinear(fin, fout, n_tasks, use_bn=False, act_init='relu')[source]#
Bases:
Module
- class models.twf_utils.afd.HardAttentionSoftmax(fin, fout, n_tasks, tau=0.6666666666666666)[source]#
Bases:
Module
- class models.twf_utils.afd.MultiTaskAFDAlternative(chw, n_tasks, cpt, clear_grad=False, use_overhaul_fd=False, lambda_diverse_loss=0.0, use_hard_softmax=True, teacher_forcing_or=False, lambda_forcing_loss=0.0, attn_mode='ch', resize_maps=False, min_resize_threshold=16)[source]#
Bases:
Module