AssertionError: AbstractConv2d Theano optimization failed

[nywang2019]

Thanks for sharing the good code.

1. I created a new virtual env and installed all components.
2. I modified keras.json as recommended. but some errors occurred when I ran, any one can help me? @ljsabc :
   ==================================== Using Theano backend. WARNING (theano.configdefaults): g++ not available, if using conda: conda install m2w64-toolchain WARNING (theano.configdefaults): g++ not detected ! Theano will be unable to execute optimized C-implementations (for both CPU and GPU) and will default to Python im plementations. Performance will be severely degraded. To remove this warning, set Theano flags cxx to an empty string. ./Arisa\Arisa_005.jpg ERROR (theano.gof.opt): Optimization failure due to: local_abstractconv_check ERROR (theano.gof.opt): node: AbstractConv2d{convdim=2, border_mode='half', subsample=(1, 1), filter_flip=True, imshp=(None, None, None, None), kshp=(24, 1, 3, 3), f ilter_dilation=(1, 1)}(Elemwise{add,no_inplace}.0, convolution2d_1_W) ERROR (theano.gof.opt): TRACEBACK: ERROR (theano.gof.opt): Traceback (most recent call last): File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\gof\opt.py", line 1982, in process_node replacements = lopt.transform(node) File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\tensor\nnet\opt.py", line 609, in local_abstractconv_check raise AssertionError( AssertionError: AbstractConv2d Theano optimization failed: there is no implementation available supporting the requested options. Did you exclude both "conv_dnn" and "conv_gemm" from the optimizer? If on GPU, is cuDNN available and does the GPU support it? If on CPU, do you have a BLAS library installed Theano can link against?

Traceback (most recent call last): File "test_mse.py", line 85, in test(sys.argv[1]) File "test_mse.py", line 65, in test out = model.predict(patch, batch_size=batch_size) File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\keras\engine\training.py", line 1216, in predict self._make_predict_function() File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\keras\engine\training.py", line 748, in _make_predict_function self.predict_function = K.function(inputs, File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\keras\backend\theano_backend.py", line 929, in function return Function(inputs, outputs, updates=updates, **kwargs) File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\keras\backend\theano_backend.py", line 912, in init self.function = theano.function(inputs, outputs, updates=updates, File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\compile\function.py", line 315, in function fn = pfunc(params=inputs, File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\compile\pfunc.py", line 483, in pfunc return orig_function(inputs, cloned_outputs, mode, File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\compile\function_module.py", line 1788, in orig_function fn = Maker(inputs, File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\compile\function_module.py", line 1474, in init optimizer_profile = optimizer(fgraph) File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\gof\opt.py", line 98, in call return self.optimize(fgraph) File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\gof\opt.py", line 87, in optimize ret = self.apply(fgraph, *args, **kwargs) File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\gof\opt.py", line 235, in apply sub_prof = optimizer.optimize(fgraph) File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\gof\opt.py", line 87, in optimize ret = self.apply(fgraph, *args, **kwargs) File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\gof\opt.py", line 2095, in apply nb += self.process_node(fgraph, node) File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\gof\opt.py", line 1985, in process_node self.failure_callback(e, self, File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\gof\opt.py", line 1881, in warn_inplace return NavigatorOptimizer.warn(exc, nav, repl_pairs, local_opt, node) File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\gof\opt.py", line 1867, in warn raise exc File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\gof\opt.py", line 1982, in process_node replacements = lopt.transform(node) File "D:\Anaconda3\envs\MangoLineExtraction\lib\site-packages\theano\tensor\nnet\opt.py", line 609, in local_abstractconv_check raise AssertionError( AssertionError: AbstractConv2d Theano optimization failed: there is no implementation available supporting the requested options. Did you exclude both "conv_dnn" and "conv_gemm" from the optimizer? If on GPU, is cuDNN available and does the GPU support it? If on CPU, do you have a BLAS library installed Theano can link against?

[ljsabc]

I guess you need to stick to CuDNN 5. It would be better if you can provide more env-related information. Also, as Theano and Keras 1.x are not supported, it's highly recommended to migrate the weight to something like tf or pytorch.

[nywang2019]

thank you@ljsabc. i searched Stackoverflow about this problem and found a solution in which it is suggested to add "optimizer = None" to file .theanorc. i did so, however a new problem occurred. when executing test_mse.py, something like "Arisa_005.jpg" appears on the screen (seems it begin to run), and then it stops there, nothing new happens. i have to stop it. so ,what is the reason? do you have some suggestions?

[ljsabc]

The first execution takes time. Please take a look at the CPU usage.

[nywang2019]

how long it takes for the first execution? I waited more than 10 minutes, and still, nothing happened. here is a screenshot:

the cpu occupation rate is about 10%.

[ljsabc]

I don't think you have fully understood the internal mechanism of code optimization.

1. g++ and nvcc is mandatory for a successful execution
2. nvcc/g++/other compilations will be run for the first time, taking around 10 mins on my testbed
3. 10% for all cores, or for a single core? Typically during compilation you will have 1 core fully occupied.
4. after the code has been compiled, with a decent GPU you can achieve <1s/1Mpixel speed.

[nywang2019]

1. you mean i should install something like g++ and nvcc? what should I do? i have installed cuda and cudnn.
2. there is 4 cores, total occupation rate is 10%
3. i added the following code into .theanorc file. "[global] floatX = float32 device = cpu optimizer = None" or "[global] floatX = float32 device = cuda optimizer = None" there is no difference, nothing changed. tell me if i am wrong.

[ljsabc]

I'm sorry but I think you need to sort it out yourself.

Attached is my keras.json and .theanorc (For Windows)

.theanorc

[global]
floatX = float32
device = gpu
 
[nvcc]
compiler_bindir=C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\bin
fastmath = True

[compile]
cxxflags = -O3 -ffast-math -ftree-loop-distribution -ftracer
mode = FAST_RUN

[dnn]
enabled = True
include_path=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\include
library_path=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\lib\x64

[lib]
cnmem = 0.7

keras.json

{
    "image_dim_ordering": "th", 
    "epsilon": 1e-07, 
    "floatx": "float32", 
    "backend": "theano"
}

[nywang2019]

i am still in trouble, but many thanks for your rapid response and help!