How to label an obstructed object to yield best results

[theshawnz]

Is it best to apply a bounding box to the largest, unobstructed view (which would be a similar approach to labeling objects at the edge of a picture where the complete object is not within the image) or is it best to apply a bounding box to the entire object (which could include some obstructions).

[EMCP]

You may want to look into more modern dataset annotation styles like COCO

http://cocodataset.org

VOC is quite limited in how it can handle this...