KangolHsu

Can you release the meta file？

2

Can you release the meta file of the model ：256_256_resfcn256_weight

why reconstructed face is connected in mouth，while original BFM has a hole in mouth？

2

why reconstructed face is connected in mouth，while original BFM has a hole in mouth？

Question about you paper [ShiftCNN]

1

Hi there I have some question about your paper “ShiftCNN: Generalized Low-Precision Architecture for Inference of Convolutional Neural Networks” 1. why do you use more then one codebook，in some of...

the converted weights

3

I use your scrip convert my float model to power of 2 but i get weights like this conv1_weight: `0.068853, -0.041312, -0.020656, -0.096395, -0.123936, 0.068853, 0.061968, 0.068853, 0.013771, 0.048197, -0.113608,...

about the value of weights

5

Here are some weights i trained my INQ model，but some value is strange，such as 0.007812, 0.007812。。。 i am sure i have already trained INQ model from ’count_\*0.7‘ to 'count_\*0.' `0.015625,...

how to handle the weight‘s sign in bit-shift arithmetic？

1

our weights’s sign include +1 ，-1，0. when using bit-shift replace multiplication sign*2^n*x -----> sign*(fixed(x)>>n) this way did not drop multiplication. how do you handle the multiplication in sign*(fixed(x)>>n)?

print "All quantization done and you can enjoy the power-of-two weights using check.py!" Have you release the script ‘check.py’？

if i want to train it on my own face data！how to prepare the cropped face？crop face by face detector or landmark detector？thanks

Can I use the CNN result to generate 3d face mesh ? How ?