AMP320_wf_400万预训练模型万能底丹，高参万能模型
高清数据集对练训练了 将近400万迭代次数，如果是一般的显卡，需要花费几个月时间才能训练不出来，已经删除inter_src 和 inter_dst文件在此基础上训练几万迭代就可以快速成型
如果你训练出来的结果像dst，不像src，第一是训练迭代次数太少，别人动辄百万迭代次数不是说着玩的
如果想高质量直播的，找被替换的人物，录个全角度视频，切脸当做dst，删除inter_dst文件，继续训练，可达到完美
具体训练步骤不在这里多说，重点的已经讲了
================= Model Summary ==================
== ==
== Model name: a320_AMP ==
== ==
== Current iteration: 3938938 ==
== ==
==--------------- Model Options ----------------==
== ==
== retraining_samples: False ==
== resolution: 320 ==
== face_type: wf ==
== models_opt_on_gpu: True ==
== ae_dims: 448 ==
== inter_dims: 1024 ==
== e_dims: 102 ==
== d_dims: 102 ==
== d_mask_dims: 22 ==
== morph_factor: 0.5 ==
== preview_mf: 1 ==
== masked_training: True ==
== eyes_prio: False ==
== mouth_prio: False ==
== uniform_yaw: False ==
== loss_function: SSIM ==
== blur_out_mask: False ==
== adabelief: True ==
== lr_dropout: n ==
== random_warp: True ==
== random_hsv_power: 0.0 ==
== random_downsample: False ==
== random_noise: False ==
== random_blur: False ==
== random_jpeg: False ==
== random_shadow: none ==
== background_power: 0.0 ==
== ct_mode: fs-aug ==
== random_color: False ==
== clipgrad: True ==
== use_fp16: False ==
== cpu_cap: 8 ==
== preview_samples: 4 ==
== force_full_preview: False ==
== lr: 5e-05 ==
== autobackup_hour: 8 ==
== session_name: rw on ==
== maximum_n_backups: 24 ==
== write_preview_history: False ==
== target_iter: 0 ==
== random_src_flip: False ==
== random_dst_flip: True ==
== batch_size: 8 ==
== gan_power: 0.0 ==
== gan_patch_size: 40 ==
== gan_dims: 16 ==
== gan_smoothing: 0.1 ==
== gan_noise: 0.0 ==
== ==
==----------------- Running On -----------------==
== ==
== Device index: 0 ==
== Name: NVIDIA A100-SXM4-80GB ==
== VRAM: 77.07GB ==
== ==
==================================================
***本内容需购买可见***
这底丹都分享出来了， 很感谢分享~ 请教，这。。我的4060ti 16g 能不能跑？
之前买过一个430w的底单一秒叠2代，好慢 跑一天好像才叠7万。。。ε=(´ο｀*)))唉 谢谢楼主分享 感谢分享 很感谢分享~ AMP是什么意思 签到领食，感谢分享 感谢分享
页:
[1]
2