Improving Berry Detection in Harvest Crates Using Mask R-CNN: Addressing Challenges with Large Image Inputs

[Ousman2001]

Hello!

I am trying to implement a feature that allows a large image to be fed into a Mask R-CNN model to detect small objects like berries in harvest crates.

Instead of directly feeding the complete image, my intention is to "split" the input image into several smaller images, each containing segmented berries, which I will then pass through the network. I have trained the model on these smaller images, and it successfully predicts the berries in these small images. However, in the images of the harvest crates, the model fails to segment the berries and only draws a bounding box around the entire crate. My question is: how can the model detect the berries in the complete image?

Thank you for your help.