CKPT_model

[ArthurConan]

Hello again, I really enjoy your program
But can you help me with such question: this tool works with ImageNet model ".pb"
But is it possible to import "model.ckpt" ?
Thanks in advance

[huanzhang12]

Yes it is possible. You can convert a "model.ckpt" to a ".pb". You can follow the "How to evaluate my own model" section in this tutorial on my other repository:
https://github.com/huanzhang12/CLEVER

[ArthurConan]

sorry for bothering you
i want to test by your ZOO attack the model "ens_adv_inception_resnet_v2_2017_08_18.tar.gz"

but i tried to use your tutorial and failed
and tried to convert by tensorflow export_inference_graph.py file and also failed
cause i get such error

File "test_all.py", line 262, in
main(args)
File "test_all.py", line 102, in main
data, model = ImageNet(), InceptionModel(sess, use_log)
File "/root/Documents/python/ZOO-Attack-master/setup_inception.py", line 225, in init
self.model = InceptionModelPrediction(sess, use_log)
File "/root/Documents/python/ZOO-Attack-master/setup_inception.py", line 192, in init
return_elements=[output_name])
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/util/deprecation.py", line 432, in new_func
return func(*args, **kwargs)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/importer.py", line 753, in import_graph_def
% ', '.join(absent_input_keys))
ValueError: Attempted to map inputs that were not found in graph_def: [input:0]

[huanzhang12]

If you are able to generate the frozen probobuf, you are very close to your goal :)

You need to change the input name in line 173, 221 and 227 in setup_inception.py and replace the "input:0" to the input tensor name that you observed in TensorBoard for you model. You also need to do the same for all the output nodes (like "InceptionV3/Predictions/Softmax:0" and "InceptionV3/Predictions/Reshape:0"). Replace them with the name of output nodes of your new graph. The current names are for inception v3 only.