jrosebr1 / bat-country Goto Github PK

I'm getting some errors when I try to change the model from the default GoogLenet. I'm probably just not understanding the API correctly.

This works:

>>> BatCountry('/caffe-master/models/bvlc_googlenet')

But this doesn't:

>>> BatCountry('/caffe-master/models/bvlc_alexnet')  
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "batcountry/batcountry.py", line 34, in __init__
    channel_swap=channels)
  File "/caffe-master/distribute/python/caffe/classifier.py", line 26, in __init__
    caffe.Net.__init__(self, model_file, pretrained_file, caffe.TEST)
RuntimeError: Could not open file /caffe-master/models/bvlc_alexnet/bvlc_googlenet.caffemodel

Are there additional arguments that need to be set when changing the model? I couldn't figure this out from the doc. In demo_bulk.py it seems like the only argument we pass is the path to the model.

Hi, I just test the Vincent van Gogh’s Wheat Field image guided using Warhol’s Marilyn Monroe. Which end_layer do you use? I have try to use inception_3a/output ~ inception_4e/output and those cannot generate the image that you posted on http://www.pyimagesearch.com/2015/07/13/generating-art-with-guided-deep-dreaming/ Thanks.

Having a spot of trouble with prepare_guide, as described in your example code.

Traceback (most recent call last):
  File "./guided-test.py", line 30, in <module>
    features = bc.prepare_guide(Image.open(guide_image), end=layer)
  File "/usr/local/lib/python2.7/dist-packages/batcountry/batcountry.py", line 185, in prepare_guide
    src.reshape(1, 3, nH, nW)
UnboundLocalError: local variable 'nH' referenced before assignment

Here are the pertinent bits of my script.

bc = BatCountry("/home/username/caffe/models/bvlc_googlenet")

input_image = "/home/username/images/monkey.jpg"
output_image = "/home/username/test-guided.jpg"
guide_image = "/home/username/guides/starry-guide.jpg"

layer = "inception_4c/output"

features = bc.prepare_guide(Image.open(guide_image), end=layer)
image = bc.dream(np.float32(
    Image.open(input_image)),
    end=layer,
    iter_n=20,
    objective_fn=BatCountry.guided_objective,
    objective_features=features,
    )

result = Image.fromarray(np.uint8(image))
result.save(output_image)
input_image = output_image

Thanks!

I love the project, can we get a command line interface to it? or if you are open to it, I or someone else can write it.

I'm trying to replicate the results from
https://www.pyimagesearch.com/2015/08/03/deep-dream-visualizing-every-layer-of-googlenet/
However, when I get near the end, it crashes, and I'm not quite sure what to do with it.
Bat Country can run single images without difficulty, but doesn't quite make it here.
To note, I am using the bvlc_googlenet model that I got when looking into deepdreamanim.
my command line is:
python visualize_layers.py --base-model /root/deepdream/caffe/models/bvlc_googlenet --image g1.jpg --output bvlc_googlenet

The crash looks like:

[INFO] processing layer inception_5b/pool 145/152
[INFO] processing layer inception_5b/pool_proj 146/152
[INFO] processing layer inception_5b/relu_pool_proj 147/152
[ERROR] cannot use layer inception_5b/relu_pool_proj
[INFO] processing layer inception_5b/output 148/152
[INFO] processing layer pool5/7x7_s1 149/152
F0621 18:21:13.746654 2164 blob.cpp:32] Check failed: shape[i] >= 0 (-3 vs. 0)
*** Check failure stack trace: ***
Aborted (core dumped)

I'd appreciate any and all help. Thank you.

issue

Bat-country dependancies are out of date - thus when I ran
pip install bat-country
I got errors when you try to run tensorflow in same env because you have just downgraded all the packages... incl protobuf.

work around

pip uninstall bat-country
pip install --upgrade (all packages that just got downgraded) 
or 
pip install tensorflow (*should* upgrade everything that needs upgrading)

I uninstalled bat-country then re-upgraded the relevant tensorflow packages everything worked again.
then

git clone https://github.com/jrosebr1/bat-country.git
python setup.py install

and it should work...