Hello to everyone,
I trained a object detection model by using Tensorflow 2.2.0 and SSD MobileNet V2 FPNLite 640x640. It is working in my conda enviroment. But I want to use this model with openCV dnn by using readNetFromTensorflow(—.pb,—.pbtxt). So I converted my saved_model.pb to frozen_graph.pb . Now I have to convert this .pb file to —.pbttxt file to use with opencv dnn. There is a script which name is tf_text_graph_ssd.py . By using this script I can create —.pbtxt with frozen_graph.pb . But my problem is this tf_text_graph_ssd just working with tensorflow1.x because a library which name is graph_transform. So I am getting error because of this. I tried a lot of things to solve this problem. Can someone help me please.
I am using:
Cuda 10.1 cuDnn 7.6.5 ubuntu 20.04 tensorflow 2.2.0 opencv 4.2.0 gpu Nvidia GTX1650
I can’t use tensorflow 1 because I have to use cuda 10.1 because of my gpu card. If I use tensorflow 1, I can’t use my gpu because tensorflow 1.15 support cuda 10.0 .