Converting Tensorflow Object Detection model to TFLite model

This is the first video in a series about running crossplatform object detection using Tensorflow Lite on iOS and Android.
In this video we will convert an object detection tensorflow model to tflite model.
Repo: github.com/Val...

[@muhammadjunaid5166]

Thanks for the wonderful explanation

[@TheCodingNotebook]

You are welcome!

[@amerbiro5411]

Excellent! Thank you so much for a good explaining I could successfully convert the model to .tflite, but when I try it in the TensorFlow demo app, the app crashes. I have not tried to use the model in VS yet. I am really looking forward to real-world problem solving by image detection. All KZbinrs just show how to detect objects but nobody shows how to use the output "what the model recognizes". Like some apps use object detection to solve some card game.

[@TheCodingNotebook]

Thanks. Well, that's up to you to find how you would like to use object detection. Usually object detection is just one block in the "image processing pipeline", for example you can pair it with object tracking: google.github.io/mediapipe/solutions/box_tracking

[@deependrabishwakarma4612]

Awesome hats off

[@TheCodingNotebook]

Thx!

[@deependrabishwakarma4612]

@@TheCodingNotebook we aspect more computer vision projects .. thank you so much

[@deependrabishwakarma4612]

i want to learn complete android computer vision, so if you have any special courses please let us know

[@TheCodingNotebook]

No. But I would recommend checking Google's MediaPipe project: google.github.io/mediapipe/

[@CodeWithImem]

thank you for your efforts .. and please i have a question .. when i am converting the graph_model into tflite .. the size i got is 198M .. while you got 12M only .. how is that happens ?

[@TheCodingNotebook]

For the exact same model that I used?

[@anvuongnguyen5496]

great video. I did exactly what you did. But I got different order of StatefulPartitioneCall. My result is: StatefulPartitionedCall 1, StatefulPartitionedCall 3, StatefulPartitionedCall 0, StatefulPartitionedCall 1. Do you know why?

[@TheCodingNotebook]

No... I am not familiar with the inner logic of the conversion...

[@fatimahussain1668]

your are using linux, i just want to create the object detection app for android can i make it without running these commands on linux?

[@TheCodingNotebook]

This video is about model conversation, for which you need Linux. If already have a model so go ahead and build the app as you like...

[@andreahernandezt]

Hi! I convert my model to .tflite the same way described here but when I import it into the app you develop in other videos it crashes because "it can't found metadata" do you had the same issue?

[@TheCodingNotebook]

Hi, no... The model I use in the example also doens't have any metadata...

[@andreahernandezt]

@@TheCodingNotebook i just figureo out that the problem wasnt with the metadata, but when i change to my .tflite model i have this issue "A/libc: Fatal signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0x4 in tid 7377 (pool-2-thread-1), pid 7307 (m.example.myapp)" it seems to be a problem with C?

[@TheCodingNotebook]

Hard to tell, probably trying to access some released memory

[@jonatanbityaev]

היי יובל אני מקבוצת רובוטיקה מתכנית first ואני ממש אשמח ליצור איתך קשר בנוגע לזה.

[@TheCodingNotebook]

Hi, there's a contact form on the website: www.thecodingnotebook.com

[@jonatanbityaev]

ok thx@@TheCodingNotebook

[@maahinpanchal5828]

I had trained my model using tensorflow and got saved_model.pb then when i converted to tflite then app is crashing.

[@TheCodingNotebook]

Try to examine the tflite model and see that the inputs/outputs shape is as you expect in the code. Another possibility is that the saved_model contains an operation that is not supported by tflite