Error in 3_runCntk.py
Opened this issue · 3 comments
Hi, I followed the sequences shown in documentation, but I faced this error when I tried to execute 3_runCntk.py. After it loaded pre-trained model, this is the error I received.
Traceback (most recent call last):
File "3_runCntk.py", line 20, in
cntk_lr_per_image, cntk_l2_reg_weight, cntk_momentum_time_constant, cntkFilesDir, boSkipTraining)
File "D:\Marcus\CNTK\ObjectDetectionUsingCntk\helpers_cntk.py", line 102, in init_train_fast_rcnn
frcn_output, frcn_penultimateLayer = frcn_predictor(image_input, roi_input, num_classes, base_path)
File "D:\Marcus\CNTK\ObjectDetectionUsingCntk\helpers_cntk.py", line 77, in frcn_predictor
roi_out = roipooling(conv_out, rois, (roi_dim, roi_dim))
File "D:\Marcus\CNTK\Anaconda3\envs\cntk-py35\lib\site-packages\cntk\internal\swig_helper.py", line 69, in wrapper
result = f(*args, **kwds)
TypeError: roipooling() missing 2 required positional arguments: 'roi_output_shape' and 'spatial_scale'
How do I generate the required arguments?
I get the same errors any clue on how to generate these arguments
Hi, what version of CNTK are you using? This might be due to breaking changes in version 2.2. Try 2.0 and this should hopefully fix this error.
Any hints on how to change the 2.2 scripts to work? I just spent hours getting CNTK set up on my machine and all the install information is now for 2.2.
Update: I reinstalled in a new python environment with 2.0 and finally it ran. Did need to update the parameters file variable svm_posweight on line 69 should to 'balanced' not 'auto'
This appears to be a change in the svm implementation.