模型里那个data.dat数据里面存储的啥东东，能打开看么？

xiaoxue

星级打分

• 1
• 2
• 3
• 4
• 5

平均分:NAN  参与人数:0  我的评分:未评

模型里那个data.dat数据里面存储的啥东东，能打开看么？

yangala

要在python里打开才看得到，正巧我早几天也好奇的看了看里面都有啥
# f = Path(r'D:\DeepFaceLab\DeepFaceLab_NVIDIA_RTX3000_series\workspace\model\mm512_SAEHD_data.dat')
# resolution 512
# face_type wf
# models_opt_on_gpu True
# archi df-ud
# ae_dims 256
# e_dims 64
# d_dims 64
# d_mask_dims 22
# masked_training True
# uniform_yaw False
# adabelief True
# lr_dropout n
# random_warp False
# gan_power 0.0
# true_face_power 0.0
# face_style_power 0.0
# bg_style_power 0.0
# ct_mode none
# clipgrad False
# pretrain False
# autobackup_hour 1
# write_preview_history False
# target_iter 1000000
# random_flip False
# batch_size 1
# eyes_mouth_prio True
# gan_patch_size 64
# gan_dims 16
# random_src_flip False
# random_dst_flip False


# 编码器
f = Path(r'D:\DeepFaceLab\DeepFaceLab_NVIDIA_RTX3000_series\workspace\model\mm512_SAEHD_encoder.npy')
# down1/downs_0/conv1/weight:0  shape =  (5, 5, 3, 64)
# down1/downs_0/conv1/bias:0  shape =  (64,)
# down1/downs_1/conv1/weight:0  shape =  (5, 5, 64, 128)
# down1/downs_1/conv1/bias:0  shape =  (128,)
# down1/downs_2/conv1/weight:0  shape =  (5, 5, 128, 256)
# down1/downs_2/conv1/bias:0  shape =  (256,)
# down1/downs_3/conv1/weight:0  shape =  (5, 5, 256, 512)
# down1/downs_3/conv1/bias:0  shape =  (512,)


# 中间层
f = Path(r'D:\DeepFaceLab\DeepFaceLab_NVIDIA_RTX3000_series\workspace\model\mm512_SAEHD_inter.npy')
# dense1/weight:0  shape =  (524288, 256)   512*512*2 = 524288
# dense1/bias:0  shape =  (256,)
# dense2/weight:0  shape =  (256, 65536)    256*256 = 65536
# dense2/bias:0  shape =  (65536,)
# upscale1/conv1/weight:0  shape =  (3, 3, 256, 1024)
# upscale1/conv1/bias:0  shape =  (1024,)


# 解码器
# f = Path(r'D:\DeepFaceLab\DeepFaceLab_NVIDIA_RTX3000_series\workspace\model\mm512_SAEHD_decoder_dst.npy')
# 18层
# upscale0/conv1/weight:0  shape =  (3, 3, 256, 2048)
# upscale0/conv1/bias:0  shape =  (2048,)
# upscale1/conv1/weight:0  shape =  (3, 3, 512, 1024)
# upscale1/conv1/bias:0  shape =  (1024,)
# upscale2/conv1/weight:0  shape =  (3, 3, 256, 512)
# upscale2/conv1/bias:0  shape =  (512,)

# res0/conv1/weight:0  shape =  (3, 3, 512, 512)
# res0/conv1/bias:0  shape =  (512,)
# res0/conv2/weight:0  shape =  (3, 3, 512, 512)
# res0/conv2/bias:0  shape =  (512,)
# res1/conv1/weight:0  shape =  (3, 3, 256, 256)
# res1/conv1/bias:0  shape =  (256,)
# res1/conv2/weight:0  shape =  (3, 3, 256, 256)
# res1/conv2/bias:0  shape =  (256,)
# res2/conv1/weight:0  shape =  (3, 3, 128, 128)
# res2/conv1/bias:0  shape =  (128,)
# res2/conv2/weight:0  shape =  (3, 3, 128, 128)
# res2/conv2/bias:0  shape =  (128,)

# out_conv/weight:0  shape =  (1, 1, 128, 3)
# out_conv/bias:0  shape =  (3,)

# upscalem0/conv1/weight:0  shape =  (3, 3, 256, 704)
# upscalem0/conv1/bias:0  shape =  (704,)
# upscalem1/conv1/weight:0  shape =  (3, 3, 176, 352)
# upscalem1/conv1/bias:0  shape =  (352,)
# upscalem2/conv1/weight:0  shape =  (3, 3, 88, 176)
# upscalem2/conv1/bias:0  shape =  (176,)

# out_convm/weight:0  shape =  (1, 1, 22, 1)
# out_convm/bias:0  shape =  (1,)
# out_conv1/weight:0  shape =  (3, 3, 128, 3)
# out_conv1/bias:0  shape =  (3,)
# out_conv2/weight:0  shape =  (3, 3, 128, 3)
# out_conv2/bias:0  shape =  (3,)
# out_conv3/weight:0  shape =  (3, 3, 128, 3)
# out_conv3/bias:0  shape =  (3,)

# upscalem3/conv1/weight:0  shape =  (3, 3, 44, 88)
# upscalem3/conv1/bias:0  shape =  (88,)





# 优化器的参数
f = Path(r'D:\DeepFaceLab\DeepFaceLab_NVIDIA_RTX3000_series\workspace\model\mm512_SAEHD_src_dst_opt.npy')
# iters:0
#
# ms_encoder/down1/downs_0/conv1/weight_0:0  shape =  (5, 5, 3, 64)
# ms_encoder/down1/downs_0/conv1/bias_0:0  shape =  (64,)
# ms_encoder/down1/downs_1/conv1/weight_0:0  shape =  (5, 5, 64, 128)
# ms_encoder/down1/downs_1/conv1/bias_0:0  shape =  (128,)
# ms_encoder/down1/downs_2/conv1/weight_0:0  shape =  (5, 5, 128, 256)
# ms_encoder/down1/downs_2/conv1/bias_0:0  shape =  (256,)
# ms_encoder/down1/downs_3/conv1/weight_0:0  shape =  (5, 5, 256, 512)
# ms_encoder/down1/downs_3/conv1/bias_0:0  shape =  (512,)

# ms_inter/dense1/weight_0:0  shape =  (524288, 256)
# ms_inter/dense1/bias_0:0  shape =  (256,)
# ms_inter/dense2/weight_0:0  shape =  (256, 65536)
# ms_inter/dense2/bias_0:0  shape =  (65536,)
# ms_inter/upscale1/conv1/weight_0:0  shape =  (3, 3, 256, 1024)
# ms_inter/upscale1/conv1/bias_0:0  shape =  (1024,)

# ms_decoder_src/upscale0/conv1/weight_0:0  shape =  (3, 3, 256, 2048)
# ms_decoder_src/upscale0/conv1/bias_0:0  shape =  (2048,)
# ms_decoder_src/upscale1/conv1/weight_0:0  shape =  (3, 3, 512, 1024)
# ms_decoder_src/upscale1/conv1/bias_0:0  shape =  (1024,)
# ms_decoder_src/upscale2/conv1/weight_0:0  shape =  (3, 3, 256, 512)
# ms_decoder_src/upscale2/conv1/bias_0:0  shape =  (512,)
# ms_decoder_src/res0/conv1/weight_0:0  shape =  (3, 3, 512, 512)
# ms_decoder_src/res0/conv1/bias_0:0  shape =  (512,)
# ms_decoder_src/res0/conv2/weight_0:0  shape =  (3, 3, 512, 512)
# ms_decoder_src/res0/conv2/bias_0:0  shape =  (512,)
# ms_decoder_src/res1/conv1/weight_0:0  shape =  (3, 3, 256, 256)
# ms_decoder_src/res1/conv1/bias_0:0  shape =  (256,)
# ms_decoder_src/res1/conv2/weight_0:0  shape =  (3, 3, 256, 256)
# ms_decoder_src/res1/conv2/bias_0:0  shape =  (256,)
# ms_decoder_src/res2/conv1/weight_0:0  shape =  (3, 3, 128, 128)
# ms_decoder_src/res2/conv1/bias_0:0  shape =  (128,)
# ms_decoder_src/res2/conv2/weight_0:0  shape =  (3, 3, 128, 128)
# ms_decoder_src/res2/conv2/bias_0:0  shape =  (128,)
# ms_decoder_src/out_conv/weight_0:0  shape =  (1, 1, 128, 3)
# ms_decoder_src/out_conv/bias_0:0  shape =  (3,)
# ms_decoder_src/upscalem0/conv1/weight_0:0  shape =  (3, 3, 256, 704)
# ms_decoder_src/upscalem0/conv1/bias_0:0  shape =  (704,)
# ms_decoder_src/upscalem1/conv1/weight_0:0  shape =  (3, 3, 176, 352)
# ms_decoder_src/upscalem1/conv1/bias_0:0  shape =  (352,)
# ms_decoder_src/upscalem2/conv1/weight_0:0  shape =  (3, 3, 88, 176)
# ms_decoder_src/upscalem2/conv1/bias_0:0  shape =  (176,)
# ms_decoder_src/out_convm/weight_0:0  shape =  (1, 1, 22, 1)
# ms_decoder_src/out_convm/bias_0:0  shape =  (1,)
# ms_decoder_src/out_conv1/weight_0:0  shape =  (3, 3, 128, 3)
# ms_decoder_src/out_conv1/bias_0:0  shape =  (3,)
# ms_decoder_src/out_conv2/weight_0:0  shape =  (3, 3, 128, 3)
# ms_decoder_src/out_conv2/bias_0:0  shape =  (3,)
# ms_decoder_src/out_conv3/weight_0:0  shape =  (3, 3, 128, 3)
# ms_decoder_src/out_conv3/bias_0:0  shape =  (3,)
# ms_decoder_src/upscalem3/conv1/weight_0:0  shape =  (3, 3, 44, 88)
# ms_decoder_src/upscalem3/conv1/bias_0:0  shape =  (88,)
# ms_decoder_dst/upscale0/conv1/weight_0:0  shape =  (3, 3, 256, 2048)
# ms_decoder_dst/upscale0/conv1/bias_0:0  shape =  (2048,)
# ms_decoder_dst/upscale1/conv1/weight_0:0  shape =  (3, 3, 512, 1024)
# ms_decoder_dst/upscale1/conv1/bias_0:0  shape =  (1024,)
# ms_decoder_dst/upscale2/conv1/weight_0:0  shape =  (3, 3, 256, 512)
# ms_decoder_dst/upscale2/conv1/bias_0:0  shape =  (512,)
# ms_decoder_dst/res0/conv1/weight_0:0  shape =  (3, 3, 512, 512)
# ms_decoder_dst/res0/conv1/bias_0:0  shape =  (512,)
# ms_decoder_dst/res0/conv2/weight_0:0  shape =  (3, 3, 512, 512)
# ms_decoder_dst/res0/conv2/bias_0:0  shape =  (512,)
# ms_decoder_dst/res1/conv1/weight_0:0  shape =  (3, 3, 256, 256)
# ms_decoder_dst/res1/conv1/bias_0:0  shape =  (256,)
# ms_decoder_dst/res1/conv2/weight_0:0  shape =  (3, 3, 256, 256)
# ms_decoder_dst/res1/conv2/bias_0:0  shape =  (256,)
# ms_decoder_dst/res2/conv1/weight_0:0  shape =  (3, 3, 128, 128)
# ms_decoder_dst/res2/conv1/bias_0:0  shape =  (128,)
# ms_decoder_dst/res2/conv2/weight_0:0  shape =  (3, 3, 128, 128)
# ms_decoder_dst/res2/conv2/bias_0:0  shape =  (128,)
# ms_decoder_dst/out_conv/weight_0:0  shape =  (1, 1, 128, 3)
# ms_decoder_dst/out_conv/bias_0:0  shape =  (3,)
# ms_decoder_dst/upscalem0/conv1/weight_0:0  shape =  (3, 3, 256, 704)
# ms_decoder_dst/upscalem0/conv1/bias_0:0  shape =  (704,)
# ms_decoder_dst/upscalem1/conv1/weight_0:0  shape =  (3, 3, 176, 352)
# ms_decoder_dst/upscalem1/conv1/bias_0:0  shape =  (352,)
# ms_decoder_dst/upscalem2/conv1/weight_0:0  shape =  (3, 3, 88, 176)
# ms_decoder_dst/upscalem2/conv1/bias_0:0  shape =  (176,)
# ms_decoder_dst/out_convm/weight_0:0  shape =  (1, 1, 22, 1)
# ms_decoder_dst/out_convm/bias_0:0  shape =  (1,)
# ms_decoder_dst/out_conv1/weight_0:0  shape =  (3, 3, 128, 3)
# ms_decoder_dst/out_conv1/bias_0:0  shape =  (3,)
# ms_decoder_dst/out_conv2/weight_0:0  shape =  (3, 3, 128, 3)
# ms_decoder_dst/out_conv2/bias_0:0  shape =  (3,)
# ms_decoder_dst/out_conv3/weight_0:0  shape =  (3, 3, 128, 3)
# ms_decoder_dst/out_conv3/bias_0:0  shape =  (3,)
# ms_decoder_dst/upscalem3/conv1/weight_0:0  shape =  (3, 3, 44, 88)
# ms_decoder_dst/upscalem3/conv1/bias_0:0  shape =  (88,)

# vs_encoder/down1/downs_0/conv1/weight_0:0  shape =  (5, 5, 3, 64)
# vs_encoder/down1/downs_0/conv1/bias_0:0  shape =  (64,)
# vs_encoder/down1/downs_1/conv1/weight_0:0  shape =  (5, 5, 64, 128)
# vs_encoder/down1/downs_1/conv1/bias_0:0  shape =  (128,)
# vs_encoder/down1/downs_2/conv1/weight_0:0  shape =  (5, 5, 128, 256)
# vs_encoder/down1/downs_2/conv1/bias_0:0  shape =  (256,)
# vs_encoder/down1/downs_3/conv1/weight_0:0  shape =  (5, 5, 256, 512)
# vs_encoder/down1/downs_3/conv1/bias_0:0  shape =  (512,)

# vs_inter/dense1/weight_0:0  shape =  (524288, 256)
# vs_inter/dense1/bias_0:0  shape =  (256,)
# vs_inter/dense2/weight_0:0  shape =  (256, 65536)
# vs_inter/dense2/bias_0:0  shape =  (65536,)
# vs_inter/upscale1/conv1/weight_0:0  shape =  (3, 3, 256, 1024)
# vs_inter/upscale1/conv1/bias_0:0  shape =  (1024,)

# vs_decoder_src/upscale0/conv1/weight_0:0  shape =  (3, 3, 256, 2048)
# vs_decoder_src/upscale0/conv1/bias_0:0  shape =  (2048,)
# vs_decoder_src/upscale1/conv1/weight_0:0  shape =  (3, 3, 512, 1024)
# vs_decoder_src/upscale1/conv1/bias_0:0  shape =  (1024,)
# vs_decoder_src/upscale2/conv1/weight_0:0  shape =  (3, 3, 256, 512)
# vs_decoder_src/upscale2/conv1/bias_0:0  shape =  (512,)
# vs_decoder_src/res0/conv1/weight_0:0  shape =  (3, 3, 512, 512)
# vs_decoder_src/res0/conv1/bias_0:0  shape =  (512,)
# vs_decoder_src/res0/conv2/weight_0:0  shape =  (3, 3, 512, 512)
# vs_decoder_src/res0/conv2/bias_0:0  shape =  (512,)
# vs_decoder_src/res1/conv1/weight_0:0  shape =  (3, 3, 256, 256)
# vs_decoder_src/res1/conv1/bias_0:0  shape =  (256,)
# vs_decoder_src/res1/conv2/weight_0:0  shape =  (3, 3, 256, 256)
# vs_decoder_src/res1/conv2/bias_0:0  shape =  (256,)
# vs_decoder_src/res2/conv1/weight_0:0  shape =  (3, 3, 128, 128)
# vs_decoder_src/res2/conv1/bias_0:0  shape =  (128,)
# vs_decoder_src/res2/conv2/weight_0:0  shape =  (3, 3, 128, 128)
# vs_decoder_src/res2/conv2/bias_0:0  shape =  (128,)
# vs_decoder_src/out_conv/weight_0:0  shape =  (1, 1, 128, 3)
# vs_decoder_src/out_conv/bias_0:0  shape =  (3,)
# vs_decoder_src/upscalem0/conv1/weight_0:0  shape =  (3, 3, 256, 704)
# vs_decoder_src/upscalem0/conv1/bias_0:0  shape =  (704,)
# vs_decoder_src/upscalem1/conv1/weight_0:0  shape =  (3, 3, 176, 352)
# vs_decoder_src/upscalem1/conv1/bias_0:0  shape =  (352,)
# vs_decoder_src/upscalem2/conv1/weight_0:0  shape =  (3, 3, 88, 176)
# vs_decoder_src/upscalem2/conv1/bias_0:0  shape =  (176,)
# vs_decoder_src/out_convm/weight_0:0  shape =  (1, 1, 22, 1)
# vs_decoder_src/out_convm/bias_0:0  shape =  (1,)
# vs_decoder_src/out_conv1/weight_0:0  shape =  (3, 3, 128, 3)
# vs_decoder_src/out_conv1/bias_0:0  shape =  (3,)
# vs_decoder_src/out_conv2/weight_0:0  shape =  (3, 3, 128, 3)
# vs_decoder_src/out_conv2/bias_0:0  shape =  (3,)
# vs_decoder_src/out_conv3/weight_0:0  shape =  (3, 3, 128, 3)
# vs_decoder_src/out_conv3/bias_0:0  shape =  (3,)
# vs_decoder_src/upscalem3/conv1/weight_0:0  shape =  (3, 3, 44, 88)
# vs_decoder_src/upscalem3/conv1/bias_0:0  shape =  (88,)
# vs_decoder_dst/upscale0/conv1/weight_0:0  shape =  (3, 3, 256, 2048)
# vs_decoder_dst/upscale0/conv1/bias_0:0  shape =  (2048,)
# vs_decoder_dst/upscale1/conv1/weight_0:0  shape =  (3, 3, 512, 1024)
# vs_decoder_dst/upscale1/conv1/bias_0:0  shape =  (1024,)
# vs_decoder_dst/upscale2/conv1/weight_0:0  shape =  (3, 3, 256, 512)
# vs_decoder_dst/upscale2/conv1/bias_0:0  shape =  (512,)
# vs_decoder_dst/res0/conv1/weight_0:0  shape =  (3, 3, 512, 512)
# vs_decoder_dst/res0/conv1/bias_0:0  shape =  (512,)
# vs_decoder_dst/res0/conv2/weight_0:0  shape =  (3, 3, 512, 512)
# vs_decoder_dst/res0/conv2/bias_0:0  shape =  (512,)
# vs_decoder_dst/res1/conv1/weight_0:0  shape =  (3, 3, 256, 256)
# vs_decoder_dst/res1/conv1/bias_0:0  shape =  (256,)
# vs_decoder_dst/res1/conv2/weight_0:0  shape =  (3, 3, 256, 256)
# vs_decoder_dst/res1/conv2/bias_0:0  shape =  (256,)
# vs_decoder_dst/res2/conv1/weight_0:0  shape =  (3, 3, 128, 128)
# vs_decoder_dst/res2/conv1/bias_0:0  shape =  (128,)
# vs_decoder_dst/res2/conv2/weight_0:0  shape =  (3, 3, 128, 128)
# vs_decoder_dst/res2/conv2/bias_0:0  shape =  (128,)
# vs_decoder_dst/out_conv/weight_0:0  shape =  (1, 1, 128, 3)
# vs_decoder_dst/out_conv/bias_0:0  shape =  (3,)
# vs_decoder_dst/upscalem0/conv1/weight_0:0  shape =  (3, 3, 256, 704)
# vs_decoder_dst/upscalem0/conv1/bias_0:0  shape =  (704,)
# vs_decoder_dst/upscalem1/conv1/weight_0:0  shape =  (3, 3, 176, 352)
# vs_decoder_dst/upscalem1/conv1/bias_0:0  shape =  (352,)
# vs_decoder_dst/upscalem2/conv1/weight_0:0  shape =  (3, 3, 88, 176)
# vs_decoder_dst/upscalem2/conv1/bias_0:0  shape =  (176,)
# vs_decoder_dst/out_convm/weight_0:0  shape =  (1, 1, 22, 1)
# vs_decoder_dst/out_convm/bias_0:0  shape =  (1,)
# vs_decoder_dst/out_conv1/weight_0:0  shape =  (3, 3, 128, 3)
# vs_decoder_dst/out_conv1/bias_0:0  shape =  (3,)
# vs_decoder_dst/out_conv2/weight_0:0  shape =  (3, 3, 128, 3)
# vs_decoder_dst/out_conv2/bias_0:0  shape =  (3,)
# vs_decoder_dst/out_conv3/weight_0:0  shape =  (3, 3, 128, 3)
# vs_decoder_dst/out_conv3/bias_0:0  shape =  (3,)
# vs_decoder_dst/upscalem3/conv1/weight_0:0  shape =  (3, 3, 44, 88)
# vs_decoder_dst/upscalem3/conv1/bias_0:0  shape =  (88,)

tangfei

数据库文件吧  不能打开

DFL8938o9

就算能打开 能看懂吗？

lknet

一些参数吧

xiaoxue

DFL8938o9 发表于 2021-9-13 17:14
就算能打开 能看懂吗？

打开乱码，想看看里面啥东西

DFL8938o9

我随便打开了一个
€}q (X
   resolutionq
cnumpy.core.multiarray
scalar
qcnumpy
dtype
qX   i4qK K
噏Rq(KX
   <qNNNJJK tqbC
  q        唓
RqX           face_typeqX
   fq
X   models_opt_on_gpuq圶   archiqX   df-udqX   ae_dimsqhhC
  q唓RqX   e_dimsqhhC\   q唓RqX   d_dimsqhhCH   q唓RqX   d_mask_dimsqhhC   q唓Rq X   masked_trainingq!圶           eyes_prioq"塜   uniform_yawq#塜
   lr_dropoutq$X
   nq%X   random_warpq&圶           gan_powerq'hhX   f8q(K K
噏)Rq*(KhNNNJJK tq+bC        q,唓-Rq.X   true_face_powerq/hh*C        q0唓1Rq2X   face_style_powerq3hh*C        q4唓5Rq6X   bg_style_powerq7hh*C        q8唓9Rq:X   ct_modeq;X   noneq<X   clipgradq=塜   pretrainq>圶   autobackup_hourq?K X   write_preview_historyq@塜   target_iterqAK X   random_flipqB圶
   batch_sizeqCKu.

DFL8938o9

虽然有些乱码 通过那些英文你基本上就知道是什么东西了

DFL8938o9

哦 弄错了 刚才那个是SAEHD_default_options.dat里面的东西

DFL8938o9

data的那个我用notepad打开都是乱码 估计需要用数据库程序打开

xiaoxue

DFL8938o9 发表于 2021-9-13 17:31
哦 弄错了 刚才那个是SAEHD_default_options.dat里面的东西

从目前的结果来看，model里的data.dat文件存放着和帧序列号相关的东东，我想确认下