unable to start in CPU mode about interactive-deep-colorization

[JSEHV]

If you change in data/colorize_image.py under def prep_net...

caffe.set_device(gpu_id)
caffe.set_mode_gpu()

to

# caffe.set_device(gpu_id)
# caffe.set_mode_gpu()
caffe.set_mode_cpu()

it will work in CPU only mode (at least for the supplied Python Notebook demos DemoGlobalHistogramTransfer.ipynb & DemoInteractiveColorization.ipynb)

[Tranvick]

@js1982
How did you make DemoGlobalHistogramTransfer.ipynb work?
Have you fixed #1 somehow?

[JSEHV]

No issue at all you have. Installed it following these 2 guidelines on Ubuntu:

OpenCV 3.2.0
http://withr.me/install-opencv-on-ubuntu-16/

Caffe
http://caffe.berkeleyvision.org/installation.html

It's running on OpenBLAS and with CPU only mode

Regarding SciKit as mentioned in your other thread:
pip install scikit-image

Worked out of the 'box'

[richzhang]

I added a --cpu_mode flag. It is rather slow in cpu mode, so you'll have to be patient when interacting with the GUI!

[blshkv]

Ok, thanks. It works, but it is really slow. It is much slower than I could imanage actually.
It uses 1 cpu only (not threaded?) and it takes 4.30 min just to load the GUI interface and 1.40 min for each and every click (regardless of the operation).

Is anything can be optimised here which would benefit GPU users as well?

[richzhang]

I'm not sure, but perhaps you can try to find how to do multi-threading with caffe?

[blshkv]

It looks like that I have to use OpenBLAS and no further code change is required. I'll look into it.

Thank you for the great work!

[mrlzc]

@blshkv
I like you use the cpu mode, but I follow this method to run the program directly down, do you have to meet this error?The error message is:

I0526 15:59:48.609108 5621 net.cpp:744] Ignoring source layer loss_log
Setting ab cluster centers in layer: pred_ab
Setting upsampling layer kernel: pred_313_us
test_imgs/mortar_pestle.jpg
scale = 2.000000
Has been killed

[blshkv]

sorry, no idea. I suggest to open a separate bug report, provide crash core, strace, system logs or anything else which could be useful to trace the problem.

[junyanz]

@mrlzc Did you compile the Caffe with the python layer support? See this post for more details.

[leoamr]

I have the same issue:

I0526 15:59:48.609108 5621 net.cpp:744] Ignoring source layer loss_log
Setting ab cluster centers in layer: pred_ab
Setting upsampling layer kernel: pred_313_us
test_imgs/mortar_pestle.jpg
scale = 2.000000
Has been killed

I re-compiled the Caffe with the python layer support, same error.

[mrlzc]

@leoamr I have solved this problem，I use the ubuntu virtual machine, I changed the memory from 4G to 8G, to solve this problem，you can try it...good luck!

[leoamr]

@mrlzc increasing the memory solved the problem for me too, thank you very much.

[blshkv]

Guys, your problem is NOT related to this bug. This bug is fixed. Please stop flooding it with your logs and open a new one.

[junyanz]

@captkkk You probably should start a separate Github issue next time. For your question, you can remove .encode('utf-8'). See this post for more details.

[captkkk]

OK, sorry for the trouble caused.