Tensorflow Object detection API on Azure


Get Tensorflow Object detection API working on Azure

Step 1:

Spin GPU VM on Azure, I provisioned Data Science Virtual Machine for Linux (Ubuntu), NC6, GPU. You can easily follow the steps here if you are new to Azure.

Step 2:

SSH to VM


Step 3:

go to notebooks folder and clone https://github.com/Azadehkhojandi/FindingFishwithTensorflowObjectdetectionAPI

$ git clone https://github.com/Azadehkhojandi/FindingFishwithTensorflowObjectdetectionAPI.git


Step 4:

If you are using Data Science Virtual Machine, It has Jupyter hub installed on the VM and you can access it by browsing

https://yourvmpublicIP:8000


Step 5:

Open installation.ipynb and run each block



Step 6:

open models/research/object_detection/fish_detection/object_detection_tutorial.ipynb notebook and run each block to see the demo.



Train Tensorflow model with your training set



Let's say you want to train your model to detect fish. We can use this video for the training.  

Step 1:

Use VOTT to extract frames and label your images 



Step 2:
After tagging all the frames export them to Pascal VOC format.

* I have already tagged and exported the result. 


Step 3:

Now you need to convert the Pascal VOC format to TF record. Open create_pascal_tf_record.ipynb notebook and run all the blocks. You should be able to see new TFrecords created for training and validation 





Step 4:

Open train.ipynb notebook and run all the blocks.


Step 5:
Open fish_detection_tutorial.ipynb to test the new model



Comments