Closed CaptainTimberTim closed 4 years ago
Here the full error message:
2020-01-28 12:53:56.242229: W tensorflow/core/framework/op_kernel.cc:1502] OP_REQUIRES failed at conv_ops.cc:477 : Invalid argument: input depth must be evenly divisible by filter depth: 3 vs 512
Traceback (most recent call last):
File "auto_pose/test/aae_retina_single_image_pose_TD.py", line 160, in
Errors may have originated from an input operation. Input Source operations connected to node Septamino_lime_small/conv2d_4/Conv2D: Septamino_lime_small/Placeholder (defined at /home/hci/.local/lib/python2.7/site-packages/auto_pose/ae/ae_factory.py:122)
Input Source operations connected to node Septamino_lime_small/conv2d_4/Conv2D: Septamino_lime_small/Placeholder (defined at /home/hci/.local/lib/python2.7/site-packages/auto_pose/ae/ae_factory.py:122)
Original stack trace for u'Septamino_lime_small/conv2d_4/Conv2D':
File "auto_pose/test/aae_retina_single_image_pose_TD.py", line 87, in
same as #66
Hi, I have a problem I can't seem to fix. If I try to use more than one AAE for pose detection the second AaE always errors with what I wrote in the title, somewhere deep in tensorflow. It does not seem to be the AAE itself that is problematic if I switch them in experiments from
experiments = ['exp_group/my_autoencoder','exp_group/Septamino_lime_small'] to experiments = ['exp_group/Septamino_lime_small', 'exp_group/my_autoencoder']
it still has an error only if the second one is detected and needs to process_pose. So it seems that the fact that it is the second one is the problem, but how... The data I give the autopose6d function seems correct as well.
This is the position where the error occurs in the code: Rs_est, ts_est = self.all_codebooks[clas_idx].auto_pose6d(self.sess, det_img, box_xywh, self._camK, 1, self.all_train_args[clas_idx], upright=self._upright)
Or am I missing something that I need to do in order to have more than one AAE?
Thanks in advance!