get expected rpn_out_class to have shape ....

[gloddream]

i used my own train data. get errors:

Num classes (including bg) = 17
Num train samples 2125
Num val samples 394
loading weights from ./model/resnet50_weights_tf_dim_ordering_tf_kernels.h5
Unable to open file (Unable to open file: name = './model/kitti_frcnn_last.hdf5', errno = 2, error message = 'no such file or directory', flags = 0, o_flags = 0)
Could not load pretrained model weights. Weights can be found in the keras application folder https://github.com/fchollet/keras/tree/master/keras/applications
Starting training
Epoch 1/3000
## Exception: Error when checking target: expected rpn_out_class to have shape (None, None, None, 9) but got array with shape (1, 38, 42, 18)

[sanyuwen]

I hava encounter this error too!
how do you solve it ?

Num classes (including bg) = 21
Num train samples 4191
Num val samples 805
loading weights from ./model/resnet50_weights_tf_dim_ordering_tf_kernels.h5
"Unable to open object (object 'bias:0' doesn't exist)"
Could not load pretrained model weights. Weights can be found in the keras application folder https://github.com/fchollet/keras/tree/master/keras/applications
Starting training
Epoch 1/3000
Exception: Error when checking target: expected rpn_out_class to have shape (None, None, None, 9) but got array with shape (1, 38, 50, 18)

I plan to train it in VOC2007 data.

[hadarpo]

One should change the rpn and cls moels architectures so that the output is concatanted to itself. the reason is that keras does not allow different sizes of y_true and y_pred. you have to change the losses accordingly

[clovermini]

@hadarpo I got the same question, but I do not particularly understand what you mean. Can you say more in detail?