We are trying to explore the identification of different seed related defects from these images. High resolution images with small objects.
How can I build a labelling pipeline that crops say, 100x100px tiles for labelling and training?
If possible, then we could crop the inference image and stitch the results of every tile’s inference. Is this the wrong approach?
What are some other ways to approach this computer vision problem?
Any learning would be very much appreciated.
Many thanks!