0

I'm working on a project to detect disease on trees. For that we are using tensor flow. I have learned how to create my own neural network and train it. I have also learned how to use an already made neural network (inception V3) and retrain it with fine tune for my purposes Now I need to learn how to detect multiple object on the image and classify it, like sheets in the tree and see if they are healthy or not and what disease it have. SoftMax can not help me with that. I have read that tensor flow object detection can help me to do it. As we will work with different inputs (infrared, RGB, HIS, multi spectral) to measure the impact of different configurations on the detection of disease, I need access to trained or retrained neural network (I know how to do that on the models that I have already made). Question is: Will tensor flow object detection let me use the models that I have already trained? Did you know a good tutorial to learn how to use it? Thanks for all!

Pedro
  • 1
  • 3
  • I cannot fully answer if you can use your existing models are not, but if you want to know how to use object detection using transfer learning, the below SO post has multiple articles that you can go through https://stackoverflow.com/questions/50554957/tensorflow-real-time-object-detection/50663881#50663881 – Srinivas Bringu Aug 20 '18 at 18:45
  • Thenx! I will check. Any way to know the full training time? just to compare training times of the same code beteen diferent computers and configurations. with timeit i goth the error NameError: name 'TimingCallback' is not defined – Pedro Aug 21 '18 at 14:24
  • training time and results depends on several factors - Number of iterations, image size, batch size, error degradation, good labelling , hardware instance (CPU, GPU) etc., - So to know, start it and run few iterations to see for yourselves. – Srinivas Bringu Aug 22 '18 at 07:25

0 Answers0