Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Not same result #14

Open
rose-jinyang opened this issue Dec 16, 2020 · 8 comments
Open

Not same result #14

rose-jinyang opened this issue Dec 16, 2020 · 8 comments
Labels
result unmatch

Comments

@rose-jinyang
Copy link

rose-jinyang commented Dec 16, 2020 •
edited
Loading

Hi
Thanks for your continuous updating.
I converted the Keras model with the latest version of your project and compared the result from the NCNN model with one from the original model.
But the results are too different from each other.

The results from the Keras model are the following:
the real face score: 0.31723082
the fake face score: 0.1382943

The results from the NCNN model are the following:
the real face score: 0.68066406
the fake face score: 0.7392578

I send the original Keras model, the converted NCNN model, and the inference scripts.
https://we.tl/t-vJESHtg4QF
Could u check on your side again?
Thanks
@MarsTechHAN
Copy link
Owner

I see you are still using the “dense_4_Softmax_blob” as the output blob, which is a bug fixed in the newer version (The version I mentioned before).
image
I try to insert the result from the model, and get same but slightly different result from yours. All the weight and activation type are correct. Maybe there is still bug in the converter, or your model is very sensitive to noise.
One thing that prevent the model from debugging is that, the model is arranged using nested sequential, which the debugger current is not able to support. Is that possible to
rewrite the model into a flatten one (you can load the model's weight by name after you rewrite the code)?

@rose-jinyang
Copy link
Author

rose-jinyang commented Dec 17, 2020 •
edited
Loading

Hi
I send the new Keras model and the converted NCNN model, and test images.
https://we.tl/t-7TqG0RQkK2
This model's input is a full image rather than a cropped face image.
The results from the Keras model are the following.

Testing for a real image ...
Score: 0.99825186
Score: 0.9731082
Score: 0.8465639
Score: 0.9993874
Score: 0.99937063

Testing for a fake image ...
Score: 1.8224884e-05
Score: 0.00020766616
Score: 0.00014449844
Score: 6.4237774e-06
Score: 3.9179995e-06
Score: 0.0070014773
Score: 0.00067165506
Score: 9.1213835e-05

This keras model works well.
A real image score > 0.5
A fake image score < 0.5

But the results from "dense_4_blob" of the converted NCNN model are the following:
Testing for a real image ...
score: 0.042266846
score: 0.030639648
score: 0.014801025
score: 0.010948181
score: 0.006668091

Testing for a fake image ...
score: 0.0049209595
score: 0.015686035
score: 0.012001038
score: 0.010871887
score: 0.041748047
score: 0.0099105835
score: 0.012832642
score: 0.00844574

So I can NOT separate the real images and the fake images.
For flattening of a nested Keras model, please check these blogs.
https://stackoverflow.com/questions/54648296/how-to-flatten-a-nested-model-keras-functional-api/54648506
https://stackoverflow.com/questions/58114504/keras-nested-models-save-and-load-weights-separately-or-view-summary-of-all-nest

@MarsTechHAN
Copy link
Owner

Fixed in 34d81db, you can try the newest version.

@rose-jinyang
Copy link
Author

thanks

@rose-jinyang
Copy link
Author

But the results from the new converted NCNN model are the following.
it seems that there is an issue yet.

Testing for a real image ...
score: 0.013702393
score: 0.77490234
score: 0.056793213
score: 0.04916382
score: 0.25073242

Testing for a fake image ...
score: 0.027145386
score: 0.63378906
score: 0.546875
score: 0.44726562
score: 0.11468506
score: 0.85839844
score: 0.2734375
score: 0.06060791

@rose-jinyang
Copy link
Author

Hi
It seems that there is an issue in my Python script for NCNN model, exactly, in loading an image data on NCNN engine.

@MarsTechHAN
Copy link
Owner

Something I just cannot understand is that, if I injected random data, I got the same result, but if I extract the input of the ncnn, and put it into the keras (vice versa), I got something different. I am not sure is there something weird in your model, or there is some implementation issue in ncnn ( or error in the conversion).

@MarsTechHAN
Copy link
Owner 

input_1
==================================
Layer Name: conv1_pad, Layer Shape: keras->(1, 225, 225, 3) ncnn->(3, 225, 225)
Max:    keras->1.000 ncnn->1.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.496 ncnn->0.496    Var: keras->0.291 ncnn->0.291
Cosine Similarity: 0.00000
Keras Feature Map:      [0.621 0.303 0.716 0.019 0.306 0.179 0.047 0.899 0.189 0.825]
Ncnn Feature Map:       [0.621 0.303 0.716 0.019 0.306 0.179 0.047 0.899 0.189 0.825]
==================================
Layer Name: conv1, Layer Shape: keras->(1, 112, 112, 32) ncnn->(32, 112, 112)
Max:    keras->2.259 ncnn->2.259    Min: keras->-2.456 ncnn->-2.456
Mean:   keras->-0.065 ncnn->-0.065  Var: keras->0.509 ncnn->0.509
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv1_bn, Layer Shape: keras->(1, 112, 112, 32) ncnn->(32, 112, 112)
Max:    keras->25.992 ncnn->25.992  Min: keras->-21.338 ncnn->-21.338
Mean:   keras->1.334 ncnn->1.334    Var: keras->2.430 ncnn->2.430
Cosine Similarity: 0.00000
Keras Feature Map:      [-0.459 -0.459 -0.459 -0.459 -0.459 -0.459 -0.459 -0.459 -0.459 -0.459]
Ncnn Feature Map:       [-0.459 -0.459 -0.459 -0.459 -0.459 -0.459 -0.459 -0.459 -0.459 -0.459]
conv1_relu_Clip
==================================
Layer Name: conv1_relu, Layer Shape: keras->(1, 112, 112, 32) ncnn->(32, 112, 112)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.663 ncnn->1.663    Var: keras->1.609 ncnn->1.609
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_dw_1, Layer Shape: keras->(1, 112, 112, 32) ncnn->(32, 112, 112)
Max:    keras->161.156 ncnn->161.156    Min: keras->-129.740 ncnn->-129.740
Mean:   keras->1.124 ncnn->1.124    Var: keras->17.405 ncnn->17.405
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_dw_1_bn, Layer Shape: keras->(1, 112, 112, 32) ncnn->(32, 112, 112)
Max:    keras->47.170 ncnn->47.170  Min: keras->-43.673 ncnn->-43.673
Mean:   keras->1.502 ncnn->1.502    Var: keras->4.015 ncnn->4.015
Cosine Similarity: 0.00000
Keras Feature Map:      [-0.455 -0.455 -0.455 -0.455 -0.455 -0.455 -0.455 -0.455 -0.455 -0.455]
Ncnn Feature Map:       [-0.455 -0.455 -0.455 -0.455 -0.455 -0.455 -0.455 -0.455 -0.455 -0.455]
conv_dw_1_relu_Clip
==================================
Layer Name: conv_dw_1_relu, Layer Shape: keras->(1, 112, 112, 32) ncnn->(32, 112, 112)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.770 ncnn->1.770    Var: keras->2.113 ncnn->2.113
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_pw_1, Layer Shape: keras->(1, 112, 112, 64) ncnn->(64, 112, 112)
Max:    keras->17.378 ncnn->17.378  Min: keras->-15.308 ncnn->-15.308
Mean:   keras->0.298 ncnn->0.298    Var: keras->3.644 ncnn->3.644
Cosine Similarity: 0.00000
Keras Feature Map:      [ 1.125 -0.298  1.059  0.712  0.414  0.573  0.39  -0.259  0.797  1.784]
Ncnn Feature Map:       [ 1.125 -0.298  1.059  0.712  0.414  0.573  0.39  -0.259  0.797  1.784]
==================================
Layer Name: conv_pw_1_bn, Layer Shape: keras->(1, 112, 112, 64) ncnn->(64, 112, 112)
Max:    keras->48.888 ncnn->48.888  Min: keras->-43.727 ncnn->-43.727
Mean:   keras->1.161 ncnn->1.161    Var: keras->5.175 ncnn->5.175
Cosine Similarity: 0.00000
Keras Feature Map:      [-3.762 -9.756 -4.041 -5.503 -6.756 -6.088 -6.859 -9.593 -5.145 -0.988]
Ncnn Feature Map:       [-3.762 -9.756 -4.041 -5.503 -6.756 -6.088 -6.859 -9.593 -5.145 -0.988]
conv_pw_1_relu_Clip
==================================
Layer Name: conv_pw_1_relu, Layer Shape: keras->(1, 112, 112, 64) ncnn->(64, 112, 112)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.925 ncnn->1.925    Var: keras->1.852 ncnn->1.852
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_pad_2, Layer Shape: keras->(1, 113, 113, 64) ncnn->(64, 113, 113)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.891 ncnn->1.891    Var: keras->1.853 ncnn->1.853
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_dw_2, Layer Shape: keras->(1, 56, 56, 64) ncnn->(64, 56, 56)
Max:    keras->72.963 ncnn->72.963  Min: keras->-68.227 ncnn->-68.227
Mean:   keras->-3.552 ncnn->-3.552  Var: keras->17.202 ncnn->17.202
Cosine Similarity: 0.00000
Keras Feature Map:      [-20.354 -26.548 -17.365 -23.186 -17.634 -22.605 -14.614 -31.992 -12.047
 -23.472]
Ncnn Feature Map:       [-20.354 -26.548 -17.365 -23.186 -17.634 -22.605 -14.614 -31.992 -12.047
 -23.472]
==================================
Layer Name: conv_dw_2_bn, Layer Shape: keras->(1, 56, 56, 64) ncnn->(64, 56, 56)
Max:    keras->31.469 ncnn->31.469  Min: keras->-27.352 ncnn->-27.352
Mean:   keras->1.226 ncnn->1.226    Var: keras->2.940 ncnn->2.940
Cosine Similarity: 0.00000
Keras Feature Map:      [-0.123 -0.716  0.164 -0.394  0.138 -0.338  0.427 -1.238  0.673 -0.421]
Ncnn Feature Map:       [-0.123 -0.716  0.164 -0.394  0.138 -0.338  0.427 -1.238  0.673 -0.421]
conv_dw_2_relu_Clip
==================================
Layer Name: conv_dw_2_relu, Layer Shape: keras->(1, 56, 56, 64) ncnn->(64, 56, 56)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.745 ncnn->1.745    Var: keras->1.644 ncnn->1.644
Cosine Similarity: 0.00000
Keras Feature Map:      [0.    0.    0.164 0.    0.138 0.    0.427 0.    0.673 0.   ]
Ncnn Feature Map:       [0.    0.    0.164 0.    0.138 0.    0.427 0.    0.673 0.   ]
==================================
Layer Name: conv_pw_2, Layer Shape: keras->(1, 56, 56, 128) ncnn->(128, 56, 56)
Max:    keras->14.726 ncnn->14.726  Min: keras->-11.284 ncnn->-11.284
Mean:   keras->0.063 ncnn->0.063    Var: keras->2.705 ncnn->2.705
Cosine Similarity: 0.00000
Keras Feature Map:      [-5.647 -5.899 -4.537 -2.833 -3.53  -4.64  -4.171 -6.195 -5.835 -5.427]
Ncnn Feature Map:       [-5.647 -5.899 -4.537 -2.833 -3.53  -4.64  -4.171 -6.195 -5.835 -5.427]
==================================
Layer Name: conv_pw_2_bn, Layer Shape: keras->(1, 56, 56, 128) ncnn->(128, 56, 56)
Max:    keras->26.681 ncnn->26.681  Min: keras->-19.834 ncnn->-19.834
Mean:   keras->0.627 ncnn->0.627    Var: keras->3.404 ncnn->3.404
Cosine Similarity: 0.00000
Keras Feature Map:      [2.394 2.192 3.282 4.646 4.088 3.199 3.575 1.954 2.243 2.57 ]
Ncnn Feature Map:       [2.394 2.192 3.282 4.646 4.088 3.199 3.575 1.954 2.243 2.57 ]
conv_pw_2_relu_Clip
==================================
Layer Name: conv_pw_2_relu, Layer Shape: keras->(1, 56, 56, 128) ncnn->(128, 56, 56)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.667 ncnn->1.667    Var: keras->1.714 ncnn->1.714
Cosine Similarity: 0.00000
Keras Feature Map:      [2.394 2.192 3.282 4.646 4.088 3.199 3.575 1.954 2.243 2.57 ]
Ncnn Feature Map:       [2.394 2.192 3.282 4.646 4.088 3.199 3.575 1.954 2.243 2.57 ]
==================================
Layer Name: conv_dw_3, Layer Shape: keras->(1, 56, 56, 128) ncnn->(128, 56, 56)
Max:    keras->63.226 ncnn->63.226  Min: keras->-65.182 ncnn->-65.182
Mean:   keras->0.169 ncnn->0.169    Var: keras->8.967 ncnn->8.967
Cosine Similarity: 0.00000
Keras Feature Map:      [-16.161  -8.931  -8.482 -24.194 -22.865 -23.837 -24.515 -10.245  -9.495
  -9.133]
Ncnn Feature Map:       [-16.161  -8.931  -8.482 -24.194 -22.865 -23.837 -24.515 -10.245  -9.495
  -9.133]
==================================
Layer Name: conv_dw_3_bn, Layer Shape: keras->(1, 56, 56, 128) ncnn->(128, 56, 56)
Max:    keras->36.687 ncnn->36.687  Min: keras->-55.052 ncnn->-55.052
Mean:   keras->0.963 ncnn->0.963    Var: keras->3.022 ncnn->3.022
Cosine Similarity: 0.00000
Keras Feature Map:      [ 0.737  2.147  2.235 -0.83  -0.571 -0.761 -0.893  1.891  2.037  2.108]
Ncnn Feature Map:       [ 0.737  2.147  2.235 -0.83  -0.571 -0.761 -0.893  1.891  2.037  2.108]
conv_dw_3_relu_Clip
==================================
Layer Name: conv_dw_3_relu, Layer Shape: keras->(1, 56, 56, 128) ncnn->(128, 56, 56)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.555 ncnn->1.555    Var: keras->1.833 ncnn->1.833
Cosine Similarity: 0.00000
Keras Feature Map:      [0.737 2.147 2.235 0.    0.    0.    0.    1.891 2.037 2.108]
Ncnn Feature Map:       [0.737 2.147 2.235 0.    0.    0.    0.    1.891 2.037 2.108]
==================================
Layer Name: conv_pw_3, Layer Shape: keras->(1, 56, 56, 128) ncnn->(128, 56, 56)
Max:    keras->12.929 ncnn->12.929  Min: keras->-12.785 ncnn->-12.785
Mean:   keras->0.351 ncnn->0.351    Var: keras->2.915 ncnn->2.915
Cosine Similarity: 0.00000
Keras Feature Map:      [ 1.031 -1.02  -0.233 -1.57  -1.378 -2.568  0.243 -0.87  -1.267 -0.691]
Ncnn Feature Map:       [ 1.031 -1.02  -0.233 -1.57  -1.378 -2.568  0.243 -0.87  -1.267 -0.691]
==================================
Layer Name: conv_pw_3_bn, Layer Shape: keras->(1, 56, 56, 128) ncnn->(128, 56, 56)
Max:    keras->44.460 ncnn->44.460  Min: keras->-17.024 ncnn->-17.024
Mean:   keras->0.217 ncnn->0.217    Var: keras->3.194 ncnn->3.194
Cosine Similarity: 0.00000
Keras Feature Map:      [ 3.414  1.212  2.057  0.622  0.828 -0.449  2.568  1.373  0.948  1.565]
Ncnn Feature Map:       [ 3.414  1.212  2.057  0.622  0.828 -0.449  2.568  1.373  0.948  1.565]
conv_pw_3_relu_Clip
==================================
Layer Name: conv_pw_3_relu, Layer Shape: keras->(1, 56, 56, 128) ncnn->(128, 56, 56)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.140 ncnn->1.140    Var: keras->1.607 ncnn->1.607
Cosine Similarity: 0.00000
Keras Feature Map:      [3.414 1.212 2.057 0.622 0.828 0.    2.568 1.373 0.948 1.565]
Ncnn Feature Map:       [3.414 1.212 2.057 0.622 0.828 0.    2.568 1.373 0.948 1.565]
==================================
Layer Name: conv_pad_4, Layer Shape: keras->(1, 57, 57, 128) ncnn->(128, 57, 57)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.100 ncnn->1.100    Var: keras->1.593 ncnn->1.593
Cosine Similarity: 0.00000
Keras Feature Map:      [3.414 1.212 2.057 0.622 0.828 0.    2.568 1.373 0.948 1.565]
Ncnn Feature Map:       [3.414 1.212 2.057 0.622 0.828 0.    2.568 1.373 0.948 1.565]
==================================
Layer Name: conv_dw_4, Layer Shape: keras->(1, 28, 28, 128) ncnn->(128, 28, 28)
Max:    keras->51.246 ncnn->51.246  Min: keras->-61.562 ncnn->-61.562
Mean:   keras->-2.560 ncnn->-2.560  Var: keras->14.385 ncnn->14.385
Cosine Similarity: 0.00000
Keras Feature Map:      [-11.011 -11.383 -16.514 -10.125 -14.622 -13.594  -6.4   -13.466 -21.025
 -15.856]
Ncnn Feature Map:       [-11.011 -11.383 -16.514 -10.125 -14.622 -13.594  -6.4   -13.466 -21.025
 -15.856]
==================================
Layer Name: conv_dw_4_bn, Layer Shape: keras->(1, 28, 28, 128) ncnn->(128, 28, 28)
Max:    keras->8.996 ncnn->8.996    Min: keras->-8.907 ncnn->-8.907
Mean:   keras->1.084 ncnn->1.084    Var: keras->1.991 ncnn->1.991
Cosine Similarity: 0.00000
Keras Feature Map:      [ 0.654  0.574 -0.521  0.843 -0.117  0.102  1.638  0.129 -1.484 -0.381]
Ncnn Feature Map:       [ 0.654  0.574 -0.521  0.843 -0.117  0.102  1.638  0.129 -1.484 -0.381]
conv_dw_4_relu_Clip
==================================
Layer Name: conv_dw_4_relu, Layer Shape: keras->(1, 28, 28, 128) ncnn->(128, 28, 28)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.393 ncnn->1.393    Var: keras->1.468 ncnn->1.468
Cosine Similarity: 0.00000
Keras Feature Map:      [0.654 0.574 0.    0.843 0.    0.102 1.638 0.129 0.    0.   ]
Ncnn Feature Map:       [0.654 0.574 0.    0.843 0.    0.102 1.638 0.129 0.    0.   ]
==================================
Layer Name: conv_pw_4, Layer Shape: keras->(1, 28, 28, 256) ncnn->(256, 28, 28)
Max:    keras->12.401 ncnn->12.401  Min: keras->-5.852 ncnn->-5.852
Mean:   keras->0.055 ncnn->0.055    Var: keras->2.072 ncnn->2.072
Cosine Similarity: 0.00000
Keras Feature Map:      [-2.877 -1.833 -2.025 -2.158 -1.653 -1.671 -1.343 -1.352 -2.798 -2.374]
Ncnn Feature Map:       [-2.877 -1.833 -2.025 -2.158 -1.653 -1.671 -1.343 -1.352 -2.798 -2.374]
==================================
Layer Name: conv_pw_4_bn, Layer Shape: keras->(1, 28, 28, 256) ncnn->(256, 28, 28)
Max:    keras->14.259 ncnn->14.259  Min: keras->-9.096 ncnn->-9.096
Mean:   keras->1.491 ncnn->1.491    Var: keras->1.801 ncnn->1.801
Cosine Similarity: 0.00000
Keras Feature Map:      [0.688 1.675 1.494 1.367 1.845 1.828 2.138 2.13  0.763 1.164]
Ncnn Feature Map:       [0.688 1.675 1.494 1.367 1.845 1.828 2.138 2.13  0.763 1.164]
conv_pw_4_relu_Clip
==================================
Layer Name: conv_pw_4_relu, Layer Shape: keras->(1, 28, 28, 256) ncnn->(256, 28, 28)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.702 ncnn->1.702    Var: keras->1.334 ncnn->1.334
Cosine Similarity: 0.00000
Keras Feature Map:      [0.688 1.675 1.494 1.367 1.845 1.828 2.138 2.13  0.763 1.164]
Ncnn Feature Map:       [0.688 1.675 1.494 1.367 1.845 1.828 2.138 2.13  0.763 1.164]
==================================
Layer Name: conv_dw_5, Layer Shape: keras->(1, 28, 28, 256) ncnn->(256, 28, 28)
Max:    keras->28.762 ncnn->28.762  Min: keras->-33.721 ncnn->-33.721
Mean:   keras->-0.953 ncnn->-0.953  Var: keras->6.296 ncnn->6.296
Cosine Similarity: 0.00000
Keras Feature Map:      [0.781 4.051 4.269 3.761 5.899 6.315 7.254 7.123 1.183 1.809]
Ncnn Feature Map:       [0.781 4.051 4.269 3.761 5.899 6.315 7.254 7.123 1.183 1.809]
==================================
Layer Name: conv_dw_5_bn, Layer Shape: keras->(1, 28, 28, 256) ncnn->(256, 28, 28)
Max:    keras->11.740 ncnn->11.740  Min: keras->-15.231 ncnn->-15.231
Mean:   keras->0.191 ncnn->0.191    Var: keras->1.602 ncnn->1.602
Cosine Similarity: 0.00000
Keras Feature Map:      [-1.159  0.685  0.808  0.521  1.728  1.962  2.492  2.418 -0.933 -0.58 ]
Ncnn Feature Map:       [-1.159  0.685  0.808  0.521  1.728  1.962  2.492  2.418 -0.933 -0.58 ]
conv_dw_5_relu_Clip
==================================
Layer Name: conv_dw_5_relu, Layer Shape: keras->(1, 28, 28, 256) ncnn->(256, 28, 28)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.700 ncnn->0.700    Var: keras->0.996 ncnn->0.996
Cosine Similarity: 0.00000
Keras Feature Map:      [0.    0.685 0.808 0.521 1.728 1.962 2.492 2.418 0.    0.   ]
Ncnn Feature Map:       [0.    0.685 0.808 0.521 1.728 1.962 2.492 2.418 0.    0.   ]
==================================
Layer Name: conv_pw_5, Layer Shape: keras->(1, 28, 28, 256) ncnn->(256, 28, 28)
Max:    keras->10.821 ncnn->10.821  Min: keras->-6.399 ncnn->-6.399
Mean:   keras->0.127 ncnn->0.127    Var: keras->1.395 ncnn->1.395
Cosine Similarity: 0.00000
Keras Feature Map:      [-1.871 -1.847 -1.886 -2.661 -3.006 -2.186 -2.248 -2.336 -2.65  -2.901]
Ncnn Feature Map:       [-1.871 -1.847 -1.886 -2.661 -3.006 -2.186 -2.248 -2.336 -2.65  -2.901]
==================================
Layer Name: conv_pw_5_bn, Layer Shape: keras->(1, 28, 28, 256) ncnn->(256, 28, 28)
Max:    keras->12.228 ncnn->12.228  Min: keras->-9.135 ncnn->-9.135
Mean:   keras->0.044 ncnn->0.044    Var: keras->1.750 ncnn->1.750
Cosine Similarity: 0.00000
Keras Feature Map:      [-0.894 -0.852 -0.921 -2.298 -2.909 -1.453 -1.564 -1.72  -2.277 -2.723]
Ncnn Feature Map:       [-0.894 -0.852 -0.921 -2.298 -2.909 -1.453 -1.564 -1.72  -2.277 -2.723]
conv_pw_5_relu_Clip
==================================
Layer Name: conv_pw_5_relu, Layer Shape: keras->(1, 28, 28, 256) ncnn->(256, 28, 28)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.720 ncnn->0.720    Var: keras->1.011 ncnn->1.011
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_pad_6, Layer Shape: keras->(1, 29, 29, 256) ncnn->(256, 29, 29)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.671 ncnn->0.671    Var: keras->0.993 ncnn->0.993
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_dw_6, Layer Shape: keras->(1, 14, 14, 256) ncnn->(256, 14, 14)
Max:    keras->40.524 ncnn->40.524  Min: keras->-37.278 ncnn->-37.278
Mean:   keras->-1.205 ncnn->-1.205  Var: keras->7.478 ncnn->7.478
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_dw_6_bn, Layer Shape: keras->(1, 14, 14, 256) ncnn->(256, 14, 14)
Max:    keras->8.268 ncnn->8.268    Min: keras->-15.184 ncnn->-15.184
Mean:   keras->1.053 ncnn->1.053    Var: keras->1.799 ncnn->1.799
Cosine Similarity: 0.00000
Keras Feature Map:      [0.014 0.014 0.014 0.014 0.014 0.014 0.014 0.014 0.014 0.014]
Ncnn Feature Map:       [0.014 0.014 0.014 0.014 0.014 0.014 0.014 0.014 0.014 0.014]
conv_dw_6_relu_Clip
==================================
Layer Name: conv_dw_6_relu, Layer Shape: keras->(1, 14, 14, 256) ncnn->(256, 14, 14)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.267 ncnn->1.267    Var: keras->1.332 ncnn->1.332
Cosine Similarity: 0.00000
Keras Feature Map:      [0.014 0.014 0.014 0.014 0.014 0.014 0.014 0.014 0.014 0.014]
Ncnn Feature Map:       [0.014 0.014 0.014 0.014 0.014 0.014 0.014 0.014 0.014 0.014]
==================================
Layer Name: conv_pw_6, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->7.966 ncnn->7.966    Min: keras->-7.162 ncnn->-7.162
Mean:   keras->-0.305 ncnn->-0.305  Var: keras->1.952 ncnn->1.952
Cosine Similarity: 0.00000
Keras Feature Map:      [-1.273 -1.487 -1.202 -1.519 -1.366 -1.764 -1.667 -1.501 -1.49  -1.614]
Ncnn Feature Map:       [-1.273 -1.487 -1.202 -1.519 -1.366 -1.764 -1.667 -1.501 -1.49  -1.614]
==================================
Layer Name: conv_pw_6_bn, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->8.043 ncnn->8.043    Min: keras->-7.961 ncnn->-7.961
Mean:   keras->0.777 ncnn->0.777    Var: keras->1.723 ncnn->1.723
Cosine Similarity: 0.00000
Keras Feature Map:      [-0.638 -0.966 -0.529 -1.016 -0.78  -1.392 -1.243 -0.988 -0.971 -1.161]
Ncnn Feature Map:       [-0.638 -0.966 -0.529 -1.016 -0.78  -1.392 -1.243 -0.988 -0.971 -1.161]
conv_pw_6_relu_Clip
==================================
Layer Name: conv_pw_6_relu, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.127 ncnn->1.127    Var: keras->1.209 ncnn->1.209
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_dw_7, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->23.925 ncnn->23.925  Min: keras->-30.512 ncnn->-30.512
Mean:   keras->-0.048 ncnn->-0.048  Var: keras->3.538 ncnn->3.538
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_dw_7_bn, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->10.783 ncnn->10.783  Min: keras->-16.061 ncnn->-16.061
Mean:   keras->-0.036 ncnn->-0.036  Var: keras->1.667 ncnn->1.667
Cosine Similarity: 0.00000
Keras Feature Map:      [0.118 0.118 0.118 0.118 0.118 0.118 0.118 0.118 0.118 0.118]
Ncnn Feature Map:       [0.118 0.118 0.118 0.118 0.118 0.118 0.118 0.118 0.118 0.118]
conv_dw_7_relu_Clip
==================================
Layer Name: conv_dw_7_relu, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.561 ncnn->0.561    Var: keras->0.927 ncnn->0.927
Cosine Similarity: 0.00000
Keras Feature Map:      [0.118 0.118 0.118 0.118 0.118 0.118 0.118 0.118 0.118 0.118]
Ncnn Feature Map:       [0.118 0.118 0.118 0.118 0.118 0.118 0.118 0.118 0.118 0.118]
==================================
Layer Name: conv_pw_7, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->8.444 ncnn->8.444    Min: keras->-7.052 ncnn->-7.052
Mean:   keras->-0.024 ncnn->-0.024  Var: keras->1.322 ncnn->1.322
Cosine Similarity: 0.00000
Keras Feature Map:      [-2.028 -1.261 -1.252 -1.324 -1.662 -0.512 -0.476 -1.346 -1.265 -1.603]
Ncnn Feature Map:       [-2.028 -1.261 -1.252 -1.324 -1.662 -0.512 -0.476 -1.346 -1.265 -1.603]
==================================
Layer Name: conv_pw_7_bn, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->9.823 ncnn->9.823    Min: keras->-6.619 ncnn->-6.619
Mean:   keras->0.577 ncnn->0.577    Var: keras->1.164 ncnn->1.164
Cosine Similarity: 0.00000
Keras Feature Map:      [-1.241 -0.103 -0.09  -0.197 -0.698  1.008  1.062 -0.229 -0.109 -0.612]
Ncnn Feature Map:       [-1.241 -0.103 -0.09  -0.197 -0.698  1.008  1.062 -0.229 -0.109 -0.612]
conv_pw_7_relu_Clip
==================================
Layer Name: conv_pw_7_relu, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.812 ncnn->0.812    Var: keras->0.837 ncnn->0.837
Cosine Similarity: 0.00000
Keras Feature Map:      [0.    0.    0.    0.    0.    1.008 1.062 0.    0.    0.   ]
Ncnn Feature Map:       [0.    0.    0.    0.    0.    1.008 1.062 0.    0.    0.   ]
==================================
Layer Name: conv_dw_8, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->32.091 ncnn->32.091  Min: keras->-18.561 ncnn->-18.561
Mean:   keras->0.206 ncnn->0.206    Var: keras->3.786 ncnn->3.786
Cosine Similarity: 0.00000
Keras Feature Map:      [ 0.     0.    -0.078 -0.26   0.361  2.74   1.899 -1.538 -0.87  -0.367]
Ncnn Feature Map:       [ 0.     0.    -0.078 -0.26   0.361  2.74   1.899 -1.538 -0.87  -0.367]
==================================
Layer Name: conv_dw_8_bn, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->7.537 ncnn->7.537    Min: keras->-13.371 ncnn->-13.371
Mean:   keras->0.250 ncnn->0.250    Var: keras->1.434 ncnn->1.434
Cosine Similarity: 0.00000
Keras Feature Map:      [-0.567 -0.567 -0.607 -0.698 -0.386  0.808  0.386 -1.339 -1.004 -0.752]
Ncnn Feature Map:       [-0.567 -0.567 -0.607 -0.698 -0.386  0.808  0.386 -1.339 -1.004 -0.752]
conv_dw_8_relu_Clip
==================================
Layer Name: conv_dw_8_relu, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.695 ncnn->0.695    Var: keras->0.922 ncnn->0.922
Cosine Similarity: 0.00000
Keras Feature Map:      [0.    0.    0.    0.    0.    0.808 0.386 0.    0.    0.   ]
Ncnn Feature Map:       [0.    0.    0.    0.    0.    0.808 0.386 0.    0.    0.   ]
==================================
Layer Name: conv_pw_8, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->7.476 ncnn->7.476    Min: keras->-5.423 ncnn->-5.423
Mean:   keras->-0.212 ncnn->-0.212  Var: keras->1.254 ncnn->1.254
Cosine Similarity: 0.00000
Keras Feature Map:      [-1.336 -0.22   0.298  0.657  0.705  0.751  0.545  0.601  0.622  0.584]
Ncnn Feature Map:       [-1.336 -0.22   0.298  0.657  0.705  0.751  0.545  0.601  0.622  0.584]
==================================
Layer Name: conv_pw_8_bn, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->6.369 ncnn->6.369    Min: keras->-7.981 ncnn->-7.981
Mean:   keras->0.275 ncnn->0.275    Var: keras->1.312 ncnn->1.312
Cosine Similarity: 0.00000
Keras Feature Map:      [-1.916 -0.185  0.62   1.178  1.253  1.324  1.003  1.091  1.124  1.064]
Ncnn Feature Map:       [-1.916 -0.185  0.62   1.178  1.253  1.324  1.003  1.091  1.124  1.064]
conv_pw_8_relu_Clip
==================================
Layer Name: conv_pw_8_relu, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.683 ncnn->0.683    Var: keras->0.802 ncnn->0.802
Cosine Similarity: 0.00000
Keras Feature Map:      [0.    0.    0.62  1.178 1.253 1.324 1.003 1.091 1.124 1.064]
Ncnn Feature Map:       [0.    0.    0.62  1.178 1.253 1.324 1.003 1.091 1.124 1.064]
==================================
Layer Name: conv_dw_9, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->20.375 ncnn->20.375  Min: keras->-19.968 ncnn->-19.968
Mean:   keras->0.372 ncnn->0.372    Var: keras->3.452 ncnn->3.452
Cosine Similarity: 0.00000
Keras Feature Map:      [ 0.    -0.193 -1.321 -2.19  -1.536 -2.054 -2.016 -2.328 -2.045 -1.691]
Ncnn Feature Map:       [ 0.    -0.193 -1.321 -2.19  -1.536 -2.054 -2.016 -2.328 -2.045 -1.691]
==================================
Layer Name: conv_dw_9_bn, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->8.764 ncnn->8.764    Min: keras->-10.032 ncnn->-10.032
Mean:   keras->0.386 ncnn->0.386    Var: keras->1.366 ncnn->1.366
Cosine Similarity: 0.00000
Keras Feature Map:      [ 0.889  0.706 -0.369 -1.198 -0.575 -1.068 -1.032 -1.329 -1.06  -0.722]
Ncnn Feature Map:       [ 0.889  0.706 -0.369 -1.198 -0.575 -1.068 -1.032 -1.329 -1.06  -0.722]
conv_dw_9_relu_Clip
==================================
Layer Name: conv_dw_9_relu, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.754 ncnn->0.754    Var: keras->0.934 ncnn->0.934
Cosine Similarity: 0.00000
Keras Feature Map:      [0.889 0.706 0.    0.    0.    0.    0.    0.    0.    0.   ]
Ncnn Feature Map:       [0.889 0.706 0.    0.    0.    0.    0.    0.    0.    0.   ]
==================================
Layer Name: conv_pw_9, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->7.063 ncnn->7.063    Min: keras->-5.465 ncnn->-5.465
Mean:   keras->0.129 ncnn->0.129    Var: keras->1.206 ncnn->1.206
Cosine Similarity: 0.00000
Keras Feature Map:      [-0.944 -2.516 -1.888 -1.46  -1.339 -1.042 -1.133 -1.092 -1.431 -1.307]
Ncnn Feature Map:       [-0.944 -2.516 -1.888 -1.46  -1.339 -1.042 -1.133 -1.092 -1.431 -1.307]
==================================
Layer Name: conv_pw_9_bn, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->9.220 ncnn->9.220    Min: keras->-10.323 ncnn->-10.323
Mean:   keras->0.201 ncnn->0.201    Var: keras->1.430 ncnn->1.430
Cosine Similarity: 0.00000
Keras Feature Map:      [ 0.284 -2.606 -1.451 -0.664 -0.442  0.104 -0.063  0.012 -0.611 -0.382]
Ncnn Feature Map:       [ 0.284 -2.606 -1.451 -0.664 -0.442  0.104 -0.063  0.012 -0.611 -0.382]
conv_pw_9_relu_Clip
==================================
Layer Name: conv_pw_9_relu, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.678 ncnn->0.678    Var: keras->0.861 ncnn->0.861
Cosine Similarity: 0.00000
Keras Feature Map:      [0.284 0.    0.    0.    0.    0.104 0.    0.012 0.    0.   ]
Ncnn Feature Map:       [0.284 0.    0.    0.    0.    0.104 0.    0.012 0.    0.   ]
==================================
Layer Name: conv_dw_10, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->24.577 ncnn->24.577  Min: keras->-34.369 ncnn->-34.369
Mean:   keras->-0.182 ncnn->-0.182  Var: keras->3.272 ncnn->3.272
Cosine Similarity: 0.00000
Keras Feature Map:      [-0.489 -0.592 -0.126 -0.357 -0.441 -0.296 -0.076 -0.141 -0.217 -0.561]
Ncnn Feature Map:       [-0.489 -0.592 -0.126 -0.357 -0.441 -0.296 -0.076 -0.141 -0.217 -0.561]
==================================
Layer Name: conv_dw_10_bn, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->11.413 ncnn->11.413  Min: keras->-10.399 ncnn->-10.399
Mean:   keras->0.455 ncnn->0.455    Var: keras->1.538 ncnn->1.538
Cosine Similarity: 0.00000
Keras Feature Map:      [1.159 1.123 1.284 1.204 1.175 1.225 1.301 1.279 1.252 1.134]
Ncnn Feature Map:       [1.159 1.123 1.284 1.204 1.175 1.225 1.301 1.279 1.252 1.134]
conv_dw_10_relu_Clip
==================================
Layer Name: conv_dw_10_relu, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.863 ncnn->0.863    Var: keras->1.044 ncnn->1.044
Cosine Similarity: 0.00000
Keras Feature Map:      [1.159 1.123 1.284 1.204 1.175 1.225 1.301 1.279 1.252 1.134]
Ncnn Feature Map:       [1.159 1.123 1.284 1.204 1.175 1.225 1.301 1.279 1.252 1.134]
==================================
Layer Name: conv_pw_10, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->8.295 ncnn->8.295    Min: keras->-6.005 ncnn->-6.005
Mean:   keras->0.363 ncnn->0.363    Var: keras->1.590 ncnn->1.590
Cosine Similarity: 0.00000
Keras Feature Map:      [ 0.586  1.352  1.112  0.742  0.921  0.312  0.335  0.594  0.487 -0.085]
Ncnn Feature Map:       [ 0.586  1.352  1.112  0.742  0.921  0.312  0.335  0.594  0.487 -0.085]
==================================
Layer Name: conv_pw_10_bn, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->8.924 ncnn->8.924    Min: keras->-5.675 ncnn->-5.675
Mean:   keras->0.077 ncnn->0.077    Var: keras->1.193 ncnn->1.193
Cosine Similarity: 0.00000
Keras Feature Map:      [2.897 3.292 3.168 2.977 3.069 2.756 2.768 2.901 2.846 2.551]
Ncnn Feature Map:       [2.897 3.292 3.168 2.977 3.069 2.756 2.768 2.901 2.846 2.551]
conv_pw_10_relu_Clip
==================================
Layer Name: conv_pw_10_relu, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.506 ncnn->0.506    Var: keras->0.767 ncnn->0.767
Cosine Similarity: 0.00000
Keras Feature Map:      [2.897 3.292 3.168 2.977 3.069 2.756 2.768 2.901 2.846 2.551]
Ncnn Feature Map:       [2.897 3.292 3.168 2.977 3.069 2.756 2.768 2.901 2.846 2.551]
==================================
Layer Name: conv_dw_11, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->23.093 ncnn->23.093  Min: keras->-21.688 ncnn->-21.688
Mean:   keras->-0.376 ncnn->-0.376  Var: keras->2.977 ncnn->2.977
Cosine Similarity: 0.00000
Keras Feature Map:      [ -9.656 -11.496 -11.164 -10.557 -10.747  -9.796  -9.778 -10.148  -9.991
  -9.12 ]
Ncnn Feature Map:       [ -9.656 -11.496 -11.164 -10.557 -10.747  -9.796  -9.778 -10.148  -9.991
  -9.12 ]
==================================
Layer Name: conv_dw_11_bn, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->13.245 ncnn->13.245  Min: keras->-11.178 ncnn->-11.178
Mean:   keras->0.548 ncnn->0.548    Var: keras->1.615 ncnn->1.615
Cosine Similarity: 0.00000
Keras Feature Map:      [-1.556 -3.06  -2.789 -2.293 -2.448 -1.671 -1.656 -1.959 -1.83  -1.118]
Ncnn Feature Map:       [-1.556 -3.06  -2.789 -2.293 -2.448 -1.671 -1.656 -1.959 -1.83  -1.118]
conv_dw_11_relu_Clip
==================================
Layer Name: conv_dw_11_relu, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.927 ncnn->0.927    Var: keras->1.176 ncnn->1.176
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_pw_11, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->11.332 ncnn->11.332  Min: keras->-8.324 ncnn->-8.324
Mean:   keras->-0.109 ncnn->-0.109  Var: keras->1.792 ncnn->1.792
Cosine Similarity: 0.00000
Keras Feature Map:      [-0.498 -1.687 -1.934 -1.633 -1.12  -1.214 -1.579 -1.344 -1.403 -1.608]
Ncnn Feature Map:       [-0.498 -1.687 -1.934 -1.633 -1.12  -1.214 -1.579 -1.344 -1.403 -1.608]
==================================
Layer Name: conv_pw_11_bn, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->13.981 ncnn->13.981  Min: keras->-9.234 ncnn->-9.234
Mean:   keras->-0.158 ncnn->-0.158  Var: keras->1.460 ncnn->1.460
Cosine Similarity: 0.00000
Keras Feature Map:      [ 0.195 -1.852 -2.276 -1.758 -0.875 -1.036 -1.665 -1.261 -1.363 -1.714]
Ncnn Feature Map:       [ 0.195 -1.852 -2.276 -1.758 -0.875 -1.036 -1.665 -1.261 -1.363 -1.714]
conv_pw_11_relu_Clip
==================================
Layer Name: conv_pw_11_relu, Layer Shape: keras->(1, 14, 14, 512) ncnn->(512, 14, 14)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.487 ncnn->0.487    Var: keras->0.879 ncnn->0.879
Cosine Similarity: 0.00000
Keras Feature Map:      [0.195 0.    0.    0.    0.    0.    0.    0.    0.    0.   ]
Ncnn Feature Map:       [0.195 0.    0.    0.    0.    0.    0.    0.    0.    0.   ]
==================================
Layer Name: conv_pad_12, Layer Shape: keras->(1, 15, 15, 512) ncnn->(512, 15, 15)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.424 ncnn->0.424    Var: keras->0.836 ncnn->0.836
Cosine Similarity: 0.00000
Keras Feature Map:      [0.195 0.    0.    0.    0.    0.    0.    0.    0.    0.   ]
Ncnn Feature Map:       [0.195 0.    0.    0.    0.    0.    0.    0.    0.    0.   ]
==================================
Layer Name: conv_dw_12, Layer Shape: keras->(1, 7, 7, 512) ncnn->(512, 7, 7)
Max:    keras->22.469 ncnn->22.469  Min: keras->-21.846 ncnn->-21.846
Mean:   keras->-0.892 ncnn->-0.892  Var: keras->3.700 ncnn->3.700
Cosine Similarity: 0.00000
Keras Feature Map:      [ -0.087   0.      0.      0.      0.     -1.075 -10.358]
Ncnn Feature Map:       [ -0.087   0.      0.      0.      0.     -1.075 -10.358]
==================================
Layer Name: conv_dw_12_bn, Layer Shape: keras->(1, 7, 7, 512) ncnn->(512, 7, 7)
Max:    keras->11.277 ncnn->11.277  Min: keras->-5.760 ncnn->-5.760
Mean:   keras->1.135 ncnn->1.135    Var: keras->1.609 ncnn->1.609
Cosine Similarity: 0.00000
Keras Feature Map:      [3.507 3.53  3.53  3.53  3.53  3.247 0.801]
Ncnn Feature Map:       [3.507 3.53  3.53  3.53  3.53  3.247 0.801]
conv_dw_12_relu_Clip
==================================
Layer Name: conv_dw_12_relu, Layer Shape: keras->(1, 7, 7, 512) ncnn->(512, 7, 7)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->1.237 ncnn->1.237    Var: keras->1.468 ncnn->1.468
Cosine Similarity: 0.00000
Keras Feature Map:      [3.507 3.53  3.53  3.53  3.53  3.247 0.801]
Ncnn Feature Map:       [3.507 3.53  3.53  3.53  3.53  3.247 0.801]
==================================
Layer Name: conv_pw_12, Layer Shape: keras->(1, 7, 7, 1024) ncnn->(1024, 7, 7)
Max:    keras->5.419 ncnn->5.419    Min: keras->-7.652 ncnn->-7.652
Mean:   keras->-0.149 ncnn->-0.149  Var: keras->1.621 ncnn->1.621
Cosine Similarity: 0.00000
Keras Feature Map:      [ 0.142  0.103  0.558  0.716  0.642 -0.647 -1.104]
Ncnn Feature Map:       [ 0.142  0.103  0.558  0.716  0.642 -0.647 -1.104]
==================================
Layer Name: conv_pw_12_bn, Layer Shape: keras->(1, 7, 7, 1024) ncnn->(1024, 7, 7)
Max:    keras->8.458 ncnn->8.458    Min: keras->-12.709 ncnn->-12.709
Mean:   keras->-0.860 ncnn->-0.860  Var: keras->1.660 ncnn->1.660
Cosine Similarity: 0.00000
Keras Feature Map:      [-1.524 -1.633 -0.351  0.094 -0.113 -3.746 -5.034]
Ncnn Feature Map:       [-1.524 -1.633 -0.351  0.094 -0.113 -3.746 -5.034]
conv_pw_12_relu_Clip
==================================
Layer Name: conv_pw_12_relu, Layer Shape: keras->(1, 7, 7, 1024) ncnn->(1024, 7, 7)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.279 ncnn->0.279    Var: keras->0.611 ncnn->0.611
Cosine Similarity: 0.00000
Keras Feature Map:      [0.    0.    0.    0.094 0.    0.    0.   ]
Ncnn Feature Map:       [0.    0.    0.    0.094 0.    0.    0.   ]
==================================
Layer Name: conv_dw_13, Layer Shape: keras->(1, 7, 7, 1024) ncnn->(1024, 7, 7)
Max:    keras->19.846 ncnn->19.846  Min: keras->-13.033 ncnn->-13.033
Mean:   keras->-0.403 ncnn->-0.403  Var: keras->1.482 ncnn->1.482
Cosine Similarity: 0.00000
Keras Feature Map:      [ 0.     0.    -0.127 -0.154 -0.203 -0.067  0.   ]
Ncnn Feature Map:       [ 0.     0.    -0.127 -0.154 -0.203 -0.067  0.   ]
==================================
Layer Name: conv_dw_13_bn, Layer Shape: keras->(1, 7, 7, 1024) ncnn->(1024, 7, 7)
Max:    keras->7.228 ncnn->7.228    Min: keras->-49.981 ncnn->-49.981
Mean:   keras->0.391 ncnn->0.391    Var: keras->1.846 ncnn->1.846
Cosine Similarity: 0.00000
Keras Feature Map:      [3.428 3.428 3.336 3.317 3.282 3.38  3.428]
Ncnn Feature Map:       [3.428 3.428 3.336 3.317 3.282 3.38  3.428]
conv_dw_13_relu_Clip
==================================
Layer Name: conv_dw_13_relu, Layer Shape: keras->(1, 7, 7, 1024) ncnn->(1024, 7, 7)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.814 ncnn->0.814    Var: keras->1.171 ncnn->1.171
Cosine Similarity: 0.00000
Keras Feature Map:      [3.428 3.428 3.336 3.317 3.282 3.38  3.428]
Ncnn Feature Map:       [3.428 3.428 3.336 3.317 3.282 3.38  3.428]
==================================
Layer Name: conv_pw_13, Layer Shape: keras->(1, 7, 7, 1024) ncnn->(1024, 7, 7)
Max:    keras->4.216 ncnn->4.216    Min: keras->-4.993 ncnn->-4.993
Mean:   keras->-0.668 ncnn->-0.668  Var: keras->1.250 ncnn->1.250
Cosine Similarity: 0.00000
Keras Feature Map:      [-0.55  -0.631 -0.854 -1.15  -0.921 -1.349 -1.346]
Ncnn Feature Map:       [-0.55  -0.631 -0.854 -1.15  -0.921 -1.349 -1.346]
==================================
Layer Name: conv_pw_13_bn, Layer Shape: keras->(1, 7, 7, 1024) ncnn->(1024, 7, 7)
Max:    keras->21.010 ncnn->21.010  Min: keras->-31.714 ncnn->-31.714
Mean:   keras->-7.717 ncnn->-7.717  Var: keras->5.685 ncnn->5.685
Cosine Similarity: 0.00000
Keras Feature Map:      [ -4.441  -5.239  -7.42  -10.318  -8.078 -12.275 -12.244]
Ncnn Feature Map:       [ -4.441  -5.239  -7.42  -10.318  -8.078 -12.275 -12.244]
conv_pw_13_relu_Clip
==================================
Layer Name: conv_pw_13_relu, Layer Shape: keras->(1, 7, 7, 1024) ncnn->(1024, 7, 7)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.228 ncnn->0.228    Var: keras->0.948 ncnn->0.948
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_pad_56, Layer Shape: keras->(1, 8, 8, 1024) ncnn->(1024, 8, 8)
Max:    keras->6.000 ncnn->6.000    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.174 ncnn->0.174    Var: keras->0.835 ncnn->0.835
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0. 0. 0. 0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0. 0. 0. 0. 0. 0.]
==================================
Layer Name: conv_dw_56, Layer Shape: keras->(1, 3, 3, 1024) ncnn->(1024, 3, 3)
Max:    keras->0.458 ncnn->0.458    Min: keras->-0.444 ncnn->-0.444
Mean:   keras->0.001 ncnn->0.001    Var: keras->0.048 ncnn->0.048
Cosine Similarity: 0.00000
Keras Feature Map:      [0. 0. 0.]
Ncnn Feature Map:       [0. 0. 0.]
==================================
Layer Name: conv_dw_56_bn, Layer Shape: keras->(1, 3, 3, 1024) ncnn->(1024, 3, 3)
Max:    keras->5.549 ncnn->5.549    Min: keras->-4.500 ncnn->-4.500
Mean:   keras->0.052 ncnn->0.052    Var: keras->0.614 ncnn->0.614
Cosine Similarity: 0.00000
Keras Feature Map:      [0.468 0.468 0.468]
Ncnn Feature Map:       [0.468 0.468 0.468]
conv_dw_56_relu_Clip
==================================
Layer Name: conv_dw_56_relu, Layer Shape: keras->(1, 3, 3, 1024) ncnn->(1024, 3, 3)
Max:    keras->5.549 ncnn->5.549    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.244 ncnn->0.244    Var: keras->0.390 ncnn->0.390
Cosine Similarity: 0.00000
Keras Feature Map:      [0.468 0.468 0.468]
Ncnn Feature Map:       [0.468 0.468 0.468]
==================================
Layer Name: conv_pw_56, Layer Shape: keras->(1, 3, 3, 2048) ncnn->(2048, 3, 3)
Max:    keras->1.674 ncnn->1.674    Min: keras->-1.647 ncnn->-1.647
Mean:   keras->-0.001 ncnn->-0.001  Var: keras->0.379 ncnn->0.379
Cosine Similarity: 0.00000
Keras Feature Map:      [0.486 0.746 0.502]
Ncnn Feature Map:       [0.486 0.746 0.502]
==================================
Layer Name: conv_pw_56_bn, Layer Shape: keras->(1, 3, 3, 2048) ncnn->(2048, 3, 3)
Max:    keras->2.571 ncnn->2.571    Min: keras->-3.136 ncnn->-3.136
Mean:   keras->-0.010 ncnn->-0.010  Var: keras->0.705 ncnn->0.705
Cosine Similarity: 0.00000
Keras Feature Map:      [-0.25   0.315 -0.214]
Ncnn Feature Map:       [-0.25   0.315 -0.214]
conv_pw_56_relu_Clip
==================================
Layer Name: conv_pw_56_relu, Layer Shape: keras->(1, 3, 3, 2048) ncnn->(2048, 3, 3)
Max:    keras->2.571 ncnn->2.571    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.276 ncnn->0.276    Var: keras->0.409 ncnn->0.409
Cosine Similarity: 0.00000
Keras Feature Map:      [0.    0.315 0.   ]
Ncnn Feature Map:       [0.    0.315 0.   ]
==================================
Layer Name: activation_1, Layer Shape: keras->(1, 3, 3, 2048) ncnn->(2048, 3, 3)
Max:    keras->2.571 ncnn->2.571    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.276 ncnn->0.276    Var: keras->0.409 ncnn->0.409
Cosine Similarity: 0.00000
Keras Feature Map:      [0.    0.315 0.   ]
Ncnn Feature Map:       [0.    0.315 0.   ]
==================================
Layer Name: flatten_1, Layer Shape: keras->(1, 18432) ncnn->(1, 1, 18432)
Max:    keras->2.571 ncnn->2.571    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.276 ncnn->0.276    Var: keras->0.409 ncnn->0.409
Cosine Similarity: 0.00000
Keras Feature Map:      [0.    0.338 0.575 0.    0.185 0.553 0.527 0.741 0.    0.   ]
Ncnn Feature Map:       [0.    0.338 0.575 0.    0.185 0.553 0.527 0.741 0.    0.   ]
Top-k:
Keras Top-k:    1692:2.571, 3268:2.400, 10790:2.374, 5788:2.358, 15408:2.313
ncnn Top-k:     1692:2.571, 3268:2.400, 10790:2.374, 5788:2.358, 15408:2.313
==================================
Layer Name: dense_2, Layer Shape: keras->(1, 12) ncnn->(1, 1, 12)
Max:    keras->4.518 ncnn->4.518    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.618 ncnn->0.618    Var: keras->1.248 ncnn->1.248
Cosine Similarity: 0.00000
Keras Feature Map:      [0.    0.    0.    1.03  0.    0.    1.101 0.    4.518 0.764]
Ncnn Feature Map:       [0.    0.    0.    1.03  0.    0.    1.101 0.    4.518 0.764]
Top-k:
Keras Top-k:    8:4.518, 6:1.101, 3:1.030, 9:0.764, 11:0.000
ncnn Top-k:     8:4.518, 6:1.101, 3:1.030, 9:0.764, 11:0.000
==================================
Layer Name: dense_3, Layer Shape: keras->(1, 8) ncnn->(1, 1, 8)
Max:    keras->2.119 ncnn->2.119    Min: keras->0.000 ncnn->0.000
Mean:   keras->0.265 ncnn->0.265    Var: keras->0.701 ncnn->0.701
Cosine Similarity: 0.00000
Keras Feature Map:      [0.    0.    0.    0.    2.119 0.    0.    0.   ]
Ncnn Feature Map:       [0.    0.    0.    0.    2.119 0.    0.    0.   ]
Top-k:
Keras Top-k:    4:2.119, 7:0.000, 6:0.000, 5:0.000, 3:0.000
ncnn Top-k:     4:2.119, 7:0.000, 6:0.000, 5:0.000, 3:0.000
==================================
Layer Name: dense_4, Layer Shape: keras->(1, 1) ncnn->(1, 1, 1)
Max:    keras->0.832 ncnn->0.832    Min: keras->0.832 ncnn->0.832
Mean:   keras->0.832 ncnn->0.832    Var: keras->0.000 ncnn->0.000
Cosine Similarity: 0.00000
Keras Feature Map:      [0.832]
Ncnn Feature Map:       [0.832]
Top-k:
Keras Top-k:    0:0.832
ncnn Top-k:     0:0.832

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
result unmatch
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@MarsTechHAN @rose-jinyang