How to automatically detect the foot length from a 3D scan image?

vision
1

I have some 3D scan images of the foot. How can I detect the foot length automatically from that image? I have the following information from the images only:

  1. Toes
  2. Metatarsals
  3. Heel
  4. weight
  5. sex
2

As a starting point, how would you as a human determine the foot length based on that data?

3

@Andrei_Cristea I can measure both length and width by using tape.

4

In that case, it sounds like you need to do something like the following flow. I’m afraid this isn’t going to be anywhere close to a step-by-step guide.

  • gather as much data as you can
  • label the data yourself, so you end up with pairs of (3d scan, actual length)
  • train a 3d CNN model (utilizing Conv3d layers) whose last layer is a Linear layer with out_features=1. perhaps something in this repo can inspire your choice of model. use data augmentations that leverage the fact that what you’re measuring is invariant to rotation (e.g. rotate the data, add some pixel noise, and so on). perhaps you can transfer-learn the feature layers from a comparable pretrained model, since you may not have quite enough data yourself to train it.

This isn’t a very helpful reply, it sounds like a challenging problem. Good luck!

5

@Andrei_Cristea unfortunately i don’t have enough data…do u know about any models that would be helpful to detect the landmarks of the foot?

6

Unfortunately I do not.