c3di/neuroscope

Neuroscope:An Explainable AI Toolbox for Semantic Segmentation and Image Classification of Convolutional Neural Nets

neuroscope's Issues

Unable to open file (file signature not found)

Dear Fei Chen,

I would like first to congratulate you on this impressive work!
Recently, I was working on reproducing your results, however, the provided .h5 models (classification and segmentation) are not working when loaded into the Neuroscope toolbox.
I checked all the .h5 models using the h5debug tool and I got (cannot open file). This debugging ensures that all the .h5 files included in the examples directory are not working.
Could you please update the examples folder by uploading a new working version of all the used models?

--
Best regards,
Abdul Karim GIZZINI

---

Postdoctoral Researcher
ETIS UMR 8051, CY Cergy Paris University, ENSEA, CNRS
6 avenue du Ponceau - 95014 CERGY PONTOISE CEDEX, France
Website: https://abdulkarimgizzini.org/
Email: [email protected]

Can't load model created by tf-keras

Hi, first of all, Thank you so much for the valuable contribution to the community,

I was wondering if there was a simple way to load a model created by tf-keras using your tool? I've tried to modify your tool to accept the tf-keras model, although my model was successfully loaded and displayed, when I make a prediction by clicking the Inspection then selecting layer and Grad-CAM, this error will appear:

tensorflow.python.framework.errors_impl.FailedPreconditionError: Error while reading resource variable conv2d_5/kernel from Container: localhost. This could mean that the variable was uninitialized. Not found: Container localhost does not exist. (Could not find resource: localhost/conv2d_5/kernel)
     [[{{node conv2d_5/Conv2D/ReadVariableOp}} = ReadVariableOp[dtype=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"](conv2d_5/kernel)]]

After fixing this issue by adding session and default graph, I was able to pass this error, however another error will arrive:
ValueError: Tensor Tensor("conv2d_11/BiasAdd:0", shape=(?, 544, 960, 2), dtype=float32) is not an element of this graph.
caused by this line:
compute_output_function = K.function(inputs=[model.native_model.input], outputs=[target_layer_output])

Do you have any idea how to overcome this kind of problem or plan to upgrade your tool in the future?
Once again, thank you for your time and consideration!

TypeError: object of type 'NoneType' has no len()

Dear @Max-ChenFei

My bachelor thesis projectis on Explainability of NNs for image classification and segmentation. For this project, one goal is to compare the different XAI techniques implemented in Neuroscope.

During implementation of Neuroscope, I am having trouble when reading the example models, as you can see in the following:

I am wondering what this error means. I assume that the error is either in the file “nx_pydot.py” or in the model file “.h5” of the respective model.

Many thanks in advance for your time and consideration.

Kind regards
Finn Schürmann

Import Network

Dear,

I was wondering how long it takes to load my own trained U-net, (.h5 file, 363 mb) into the toolbox?
It currently seems to be stuck at loading the file for me, even if I let it load for a couple of hours. Is there something I could do to speed it up?
I have no loading problems with the included examples.

Thanks in advance.

Resolve Package No tFound:

based on gpu.yml I want to create an environment what should I do? do you have any plan to update yml file?

Resolve Package Not Found:

• grpcio==1.23.0=py36h3db2c7e_0
• libtiff==4.1.0=h21b02b4_3
• protobuf==3.11.2=py36he025d50_0
• m2w64-gcc-libs-core==5.3.0=7
• tensorflow==1.13.1=h21ff451_8
• icu==64.2=he025d50_1
• xz==5.2.4=h2fa13f4_1001
• jpeg==9c=hfa6e2cd_1001
• tensorflow-base==1.13.1=py36_8
• pillow==7.0.0=py36h9ea1dd6_0
• kiwisolver==1.1.0=py36he980bc4_0
• libxslt==1.1.33=heafd4d3_0
• libiconv==1.15=hfa6e2cd_1005
• pyreadline==2.1=py36_1001
• m2w64-gmp==6.1.0=2
• tk==8.6.10=hfa6e2cd_0
• markupsafe==1.1.1=py36hfa6e2cd_0
• keras==2.3.1=py36h21ff451_0
• ninja==1.9.0=h1ad3211_1
• libpng==1.6.37=h7602738_0
• lz4-c==1.8.3=he025d50_1001
• graphviz==2.38.0=h6538335_1011
• scipy==1.3.1=py36h29ff71c_0
• matplotlib-base==3.1.2=py36h2981e6d_1
• m2w64-libwinpthread-git==5.0.0.4634.697f757=2
• intel-openmp==2019.5=281
• qt==5.12.5=h7ef1ec2_0
• zstd==1.4.4=hd8a0e53_1
• libclang==9.0.1=default_hf44288c_0
• pyqt==5.12.3=py36h6538335_1
• cffi==1.13.2=py36hb32ad35_0
• pygpu==0.7.6=py36hc8d92b1_1000
• h5py==2.10.0=nompi_py36h422b98e_101
• msys2-conda-epoch==20160418=1
• yaml==0.2.2=hfa6e2cd_1
• icc_rt==2019.0.0=h0cc432a_1
• python==3.6.8=h9f7ef89_0
• freetype==2.10.0=h563cfd7_1
• libprotobuf==3.11.2=h1a1b453_0
• mkl-service==2.3.0=py36hfa6e2cd_0
• libxml2==2.9.10=h9ce36c8_0
• m2w64-gcc-libs==5.3.0=7
• pyyaml==5.2=py36hfa6e2cd_0
• hdf5==1.10.5=nompi_ha405e13_1104
• pyside2==5.13.1=py36hfa7ce6d_6
• tornado==6.0.3=py36hfa6e2cd_0
• zlib==1.2.11=h2fa13f4_1006
• openssl==1.1.1d=hfa6e2cd_0
• vc=14.1
• sqlite==3.30.1=hfa6e2cd_0
• libgpuarray==0.7.6=hfa6e2cd_1003
• m2w64-gcc-libgfortran==5.3.0=6
• numpy==1.17.3=py36hc71023c_0
• pytorch==1.3.1=py3.6_cuda101_cudnn7_0
• vs2015_runtime
• wincertstore==0.2=py36_1003
• tensorflow-estimator==1.13.0=py36h39e3cac_0

Import Network doesn't work, infinite load.

Hi! I am trying to load my model to the executable, with architectures like VGG16 and VGG19 and it does not finish loading the model. Is there any solution?
With the executable, it does load the examples, but it does not finish loading the model that I select locally.