TensorFlow Lite Model in Raspberry Pi

Please help me to convert .pb file to .tflite format. I can’t convert again after new update in TensorFlow


Some more info? Version of TensorFlow that conversion worked and code snippet?


I am using Google Collab. This is My Code. Please help me in Section 12. I fail to convert .pb file to .tflite. https://colab.research.google.com/drive/1D7WrDIvuXmrQStHnXBJHBzQlo-1S8UvQ

It needs access. Please add Removed by Moderator

Done. You can access

So what is the error? I do not see executed cells to see what is wrong.

I seems that it is a standard procedure with scripts that have been created and tested from Google like “export_tflite_graph_tf2.py”. Because executing all the cells it is time consuming please train the model and use “export_tflite_graph_tf2.py” script to export to tflite file… Or train the model and execute the command to convert Saved Model to TFLite with the TensorFlow Lite converter to see the error logcat.

It will be also helpful if you create the Saved Model folder with all the contents and share it. This way we can reproduce the issue any time.

Hi, May I know which model you are using? Classification or detection? and what architecture ? I have faced similar issue. And when I tried to look into you code, it was saying access is denied

There is no error but .tflite model become file with size only 2 kb. But, file .pb size is 11 MB. When I load .tflite it will become error.

Detection, SSD-MobileNetV2. What is your e-mail? I can give you an access

Please create saved model, zip it and share it to reproduce

Where I have to share?

Upload somewhere and share the link here.

Check your inbox for the mail id

Hello Please help me why file .tflite is always become 2 Kb?
Here is the saved_model.zip saved_model.zip - Google Drive

Hi @Aulya_Yarzuki ,

Saved model file is succescfully converted to .tflite file. I have used a simple colab notebook and method from here. The file that is produced is about 10MB and is succesfully loaded in netron showing that is an object detection application that you want to build.

Check below code:

Ping me if you want anything else.


1 Like