iml130 / nncg Goto Github PK

This is very interesting work but I am not able to get this running with Tensor flow 2.5 and Kera 2.5.0.dev2021032900. Is their a combination of TF and Keras that works with nncg? I have installed python 3.7 and clang 6.
Thanks,
-gits

Ran tests using the code "applications/tests.py".

Looks like VGG16_test() test is failing in evaluation.

.....
######################################################################
strides passed
######################################################################

2021-08-30 15:52:07.957106: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 411041792 exceeds 10% of free system memory.
2021-08-30 15:52:08.068794: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 411041792 exceeds 10% of free system memory.
2021-08-30 15:52:08.141630: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 411041792 exceeds 10% of free system memory.
Finished |████████████████████████████████████████| 100.0% 7.1% ---| 0.0%
Check of variable x_13 for layer block3_conv1.
Largest error 2.3889937210419543e-07 at (31, 11, 94) (447326).
Values: 0.1227651908993721 and 0.122764952

VGG19_test() as well -

....
....
....
2021-08-30 16:07:31.739931: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 411041792 exceeds 10% of free system memory.
2021-08-30 16:07:31.860670: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 411041792 exceeds 10% of free system memory.
2021-08-30 16:07:31.937269: W tensorflow/core/framework/cpu_allocator_impl.cc:80] Allocation of 411041792 exceeds 10% of free system memory.
Finished |████████████████████████████████████████| 100.0% 7.1% ---| 0.0%
Check of variable x_15 for layer block3_conv2.
Largest error 2.2351569518852532e-07 at (9, 29, 124) (136572).
Values: 0.06643339991569519 and 0.0664331764

The quantized ones with real images are also failing since probably they use the same generated C code.
I don't get a print_success if I comment out VGG16_test() and VGG19_test() and let run VGG16_quantized_test(db), VGG19_quantized_test(db)

Hi, On using a batch normalization layer I get "Warning: BatchNormalization not implemented" but the paper mentions about using this layer in Table 3 wrt the testing done on Nao robot CNN. How can I add this layer in code?
Thanks.