How can we get the best custom model for Object Detection

We are working on a kotlin App for Object Detection wherein we are using mobilenetV2_I320.tflite model which we train from MediaPipeModel maker.

We have been facing the problem in terms of accuracy in detection. Can anyone please help us with below points.

  1. What are the things we should do before Annotations ?
  2. What are the steps we must apply in Preprocessing ?
  3. What should we do with model training. We are exporting data in Pascal VOC format then we start model building in Google Colab using MediaPipeModel Maker where we have the limitations of GPUs, it requires min L4 GPU which is paid. Here as per their steps, we do 50 epoch on the learning rate of 0.3 and then we export the model to use it offline.
  4. Is there any Other options we can go with ? where we can get offline models which are trained from our custom data ?

Please help, This can save my life.

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.