What could be the best model to tackle this task?

  • **Project Type: Box and bunlde counting
  • **Operating System & Browser: Windows/ Linux, Brave or Edge
  • **Project Universe Link or Workspace/Project ID: Working on it

Hello to everyone, my name is Andres, Im a cs engineer, currently working on paperboard industry, I would like to know if anybody has worked with this type of task.

The goal is to count the boxes but on a more grained way, so the count is more precise, this is because at the moment they are counted by hand. The machine has an integrated counting sensor, so the idea is to compare both and be more efficient.

Hi @Jose_Andres_Millan!
This is a great project and a fantastic use case for Roboflow!! Given your example image (tightly stacked boxes), I would suggest implementing a key point detection model where you place key points on the visible edge, stripe, or fold. You may have to tinker with these key points to see which works best.

Here is our documentation on implementing Keypoint Detection. Additionally, here are our resources on annotating Keypoints.

Once you have a model trained, here is an example workflow:

Blocks to note:

  • Dataset Upload: Allows you to review predictions and add input to your dataset. Powerful tool to help improve your model.
  • KeyPoint Visualization: Block that I like to use to help with quick checks. Overlays keypoint predictions onto inputted image for visual debugging.

Please keep us updated on your progress!!