search

ahmetozlu / tensorflow_object_counting_api

🚀 The TensorFlow Object Counting API is an open source framework built on top of TensorFlow and Keras that makes it easy to develop object counting systems!
https://www.youtube.com/watch?v=yT_1eKJTdfk
MIT License
1.32k stars 545 forks source link

Problem trying to load different model not in TF ZOO #77

Closed sagarkar10 closed 4 years ago

sagarkar10 commented 4 years ago

I am trying to use the counting API but the backbone I want to use is EfficientDet here

I tried changing the forzen_graph but it will not work due to different arch. The tensor names are obviously not same, what is the way to proceed towards it. In fact, is it the appropriate way to approach this problem. I want the Counting API on models like EfficientDet/YOLO

Little help will be very much appreciated.

sagarkar10 commented 4 years ago

@ahmetozlu please help

ahmetozlu commented 4 years ago

As long as the frozen graph (.pb) is located under custom_frozen_inference_graph and the label map is located under data, the counting API runs to detect & track & count the objects using the custom frozen graph you want to use!

sagarkar10 commented 4 years ago

But the tensor names and tensor structures aren't same. How do I tackle that?

aparico commented 3 years ago

As long as the frozen graph (.pb) is located under custom_frozen_inference_graph and the label map is located under data, the counting API runs to detect & track & count the objects using the custom frozen graph you want to use!

@sagarkar10 Did the suggestion work?