OpenCV DNN Throws Unsupported Layer Error

asked 2019-11-16 02:24:14 -0500

aditya_mangalampalli gravatar image

Hello all, I recently retrained a model and wanted to run it using OpenCV's DNN module. However, when I try to run the file, I keep getting this error.

Traceback (most recent call last): File "", line 23, in <module> net = cv2.dnn.readNetFromTensorflow('/Users/Robot/Skystone-Vision/SkystoneVision/PB-File/model.pb') cv2.error: OpenCV(4.1.2-dev) /Users/Robot/opencv/modules/dnn/src/tensorflow/tf_graph_simplifier.cpp:90: error: (-212:Parsing error) Input node with name Squeeze not found in function 'getInputNodeId'

I do not know how to proceed next as I have no clue on how to fix this problem. Please help if you can. I will be attaching my checkpoint files below.

Any help would be greatly appreciated. Thanks!

edit retag flag offensive close merge delete


what kind of model did you try to retrain ? can you link us to the training code or such ?

(the tf importer does support Squeeze layers, it rather looks like a broken connection or a naming problem)

berak gravatar imageberak ( 2019-11-16 04:41:11 -0500 )edit

@berak, I retrained a Mobilenet V2 with SSDLite. I didn't exactly use a code for training the model but I can provide the checkpoints which was the google drive folder. I used for the object detection training which also means thats I do not have a proper code for how the model is trained, though I can dig around the github some more and find out. The link has been updated to show my inferencing script.

aditya_mangalampalli gravatar imageaditya_mangalampalli ( 2019-11-16 09:54:14 -0500 )edit