Is Detectron2 the right choice for multi label classification problem?

[Aayushjshah]

The project is a multi label classification problem. I have selected Detectron2 as the base model.Currently we are using the instance segmentation model of detectron2 based on FPN(COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x).
The classification is sensitive to shades of colors of the object in the images.

Questions)
1)Is detectron2 a right choice of model for multi label classification of images ?
2) Currently , we are achieving 60% accuracy. How to fine tune model to improve accuracy?
3) Also , in your expertise is the Instance segmentation model good for the problem?If not , can you suggest another model which will be more suitable.
4) Also we are currently using the default predictor itself.Can we use some other predictor which would be more suitable for the problem.

Any further advice/suggestion on the problem is appreciated.
Thanking you in advance for helping me out :) !!