ckmarkoh / neuralart_tensorflow Goto Github PK

I would like to use this in a mobile application, and the best way I see is to use it with Firebase ML kit.
The kit only works with tensorFlow Lite models https://www.tensorflow.org/mobile/tflite/

Please can you convert, or tell me how to properly convert your model to a the Tensorflow lite model?

When i run main.py, it show some error

Traceback (most recent call last):
File "main.py", line 144, in
main()
File "main.py", line 112, in main
content_img = read_image(CONTENT_IMG)
File "main.py", line 96, in read_image
image = image[np.newaxis,:IMAGE_H,:IMAGE_W,:]
IndexError: too many indices for array

how do i fix it, Thx

Only the first image (0000.png) has any content - a slightly artsy-fied version of the original. It looks like it works fine for the first image. But everything else after that is black frames; those are not corrupted images, just solid black frames, about 1.5 kB each.

numpy (1.11.1)
scipy (0.18.0)
tensorflow (0.10.0rc0) - installed from binary
Pillow (3.3.1)
CUDA 7.5.18
cuDNN 4
Nvidia driver 367.44
Ubuntu 14.04
kernel 4.4.0-36
Nvidia Titan X Pascal

Here's the output from the script:

$ python main.py 
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcublas.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcudnn.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcufft.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcurand.so locally
I tensorflow/core/common_runtime/gpu/gpu_init.cc:102] Found device 0 with properties: 
name: TITAN X (Pascal)
major: 6 minor: 1 memoryClockRate (GHz) 1.531
pciBusID 0000:01:00.0
Total memory: 11.90GiB
Free memory: 11.48GiB
I tensorflow/core/common_runtime/gpu/gpu_init.cc:126] DMA: 0 
I tensorflow/core/common_runtime/gpu/gpu_init.cc:136] 0:   Y 
I tensorflow/core/common_runtime/gpu/gpu_device.cc:839] Creating TensorFlow device (/gpu:0) -> (device: 0, name: TITAN X (Pascal), pci bus id: 0000:01:00.0)
6.48804e+11
9.17615e+11
9.17613e+11
9.17614e+11
nan
9.17614e+11
inf
nan
nan
nan
9.1761e+11
4.88761e+21
nan
inf
nan
nan
inf
nan
nan
nan
nan
9.17614e+11
nan
nan
nan
nan
nan
nan
7.94981e+18
nan
2.32385e+21
nan
nan
3.28392e+21
inf
nan
nan
1.3813e+12
4.89096e+21
inf
nan
nan
nan
nan
nan
nan
inf
nan
nan
nan

"assert layer_name == expected_layer_name" failed when I used the VGG-19 model linked in the readme. It was 534.9MB and did not seem to have proper layer names. I used the model at https://github.com/ckmarkoh/neuralart_tensorflow (which is in the readm2). Files is at https://drive.google.com/file/d/0B8QJdgMvQDrVU2cyZjFKU1RrLUU/view
it was 576mb and worked fine.

p.s. Thank for the notebook, easy to understand.

ubuntu16
tf 1.0
is this api it changed?

Hi Mark, I have read the paper, and watched your video, I think I got the very basic idea of the style transfer. However, there is one line I don't understand (maybe I am really not familiar with tensrorflow)
In the following
cost_content = sum(map(lambda l,: l[1]*build_content_loss(sess.run(net[l[0]]) , net[l[0]])
, CONTENT_LAYERS))

why using sess.run(net[l[0]]) vs. net[l[0] as cost computation ? what is the difference for net[0] in "sess.run' vs net[0] ?