Larger masks than 28x28

[Woz4tetra]

I've had some good initial success with mask RCNN. I trained from this model config mask_rcnn_R_50_FPN_3x on a custom dataset in order to find the robot playing field. However, I've noticed the edges are quite wavy. I trained on 4852 images (1620 original images. I applied a random augmentation to each image twice). I think I have enough data. I suspect the object I'm trying to locate is just too large in relation to the rest of the image. The 28x28 mask gets stretched and the fuzzy edges result in these very wavy lines. Is there a way to increase the mask size? Alternatively is there a way to better smooth these edges? I'm using this function to convert the masks into contours: from detectron2.layers import paste_masks_in_image.