Models: TypeError: `pred` must be a Tensor, or a Python bool, or 1 or 0. Found instead: None

[dtuyenle]

Keep having below issue when run object_detection/train.py
TypeError: pred must be a Tensor, or a Python bool, or 1 or 0. Found instead: None

[nickyua]

change 109 line in ssd_mobilenet_v1_feature_extractor.py:
from

              is_training=None, regularize_depthwise=True)):

to

              is_training=True, regularize_depthwise=True)):

[dtuyenle]

that works. thanks

[hakillha]

@nickyua I think this can potentially cause serious problems since the export_inference_graph.py script runs this module to build the models as well.
I found this problem since I somehow got to export the frozen graphs with the older and the newer version of the API and the graphs differed in performance dramatically. So I changed that line to is_training=False when I was exporting the graph with the newer API which fixed the problem.

[pkulzc]

This problem exists for a certain period of time when slim dir is outdated. Syncing to latest and this should be gone.

[leccyril]

where is ssd_mobilenet_v1_feature_extractor.py ?

[tesfagabir]

@pkulzc I still have the same problem even if I update the tensorflow/models to its latest version. I have a version which was updated on July 17. But still getting the problem. I also changed from training=None to training=True on ssd_mobilenet_v2_feature_extractor.py module. Could you please help me on that?

[pkulzc]

Are you still setting training=True? You should never hardcode training because it may keep updating batch norm when running eval.

[tesfagabir]

Yes I did. So should I bring it back to training=None?

[pkulzc]

Yes, you should revert your change.

On Sat, Aug 11, 2018 at 10:08 AM Tesfagabir Meharizghi <
[email protected]> wrote:

Yes I did. So should I bring it back to training=None?

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/tensorflow/models/issues/4043#issuecomment-412288609,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AiMUOIjHi0Hxo7LQY6Tt0s--wq8AkJNCks5uPw-FgaJpZM4TeWII
.

[Elites2017]

0
down vote
favorite

I want to train my custom model using the pretained model " mobilenet_v1_0.5_128 " / " mobilenet_v1_1.0_224 " from tensorflow downloaded at: download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_224.tgz which will make it easier for me to convert it to tflite later

Where can I find the right config file matched to those models (or to one of them)? Because when I use some config files from the tensorflow github official repo, when lauching the trainning I got the followig error: ValueError: No variables to save

I'm sure the probleme is because I don't use the correct config file.

Thanks in advance for your help guys !

[florinmatei]

where is ssd_mobilenet_v1_feature_extractor.py ?

I can't find this file either. Were you able to find it?