Roadway asset libary creation

Hello everyone, I have Right-of-way images as you drive down the road, and want to train DS to find things along the roadway… if possible I want to teach it to find extra turning lanes… sign posts (some day sign faces according to my libary).

My question, would anyone have a libary or “how to” process already built? I have 75,000 pictures of sign faces and posts taken with hand held cameras and close to 5 million van collected images I want to run the AI on. Just want to see if this is a possible solution.

1 Like

@TrevorMillen This is definitely a possible solution. Based on your post, I will suggest the following steps going forward

  • have a list of specific items you want to train DeepStack on
    • Are you detecting sign posts generally
    • Are you detecting some specific sign posts/faces ( by size, color, brand, type)
  • Once you have the list of items you want to detect, then set aside 1000-2000 instances of each of the items from the 75,000 that you have
  • Follow the instructions detailed for training custom DeepStack detection models https://docs.deepstack.cc/custom-models to annotate train and deploy the model on the other 5 million images.
  • Extra notes:
    • Annotate the images for each of objects with LabelIMG (details in the link above). This is what will take the most time for you but it is worth it.
    • Train a yolov5x model using the sample Colab notebook provided in the link above
    • Deploy and test on samples from the 5 million images.