Can U-Net and segmentation be used for predicting objects placement on image?

Let’s say I have dataset of several thousand images of polygons with some objects placed on the polygon edges. If I create a mask for training that will show where said objects are on each image, and then train U-Net or some other CNN so it will be able to find these objects on new images, can I also use this to predict optimal (based on dataset) placement of objects on clean polygons (that doesn’t have objects on them whatsoever)?

I am taking this approach to create an AI that can place sprinklers on lawn, so I am interested if this solution is possible, or I skipped some easier way to do this.

1 Like

I my opinion CNNs and U Net help to detect the things, if the objects are always placed in the same fashion on the polygon then try to classify the images with and without the objects and then design your segmentation map on the polygon for the objects to be placed.

I never tried such a thing but would love to give it a shot if you could send me the link to the dataset.