train a model without axis aligned bounding boxes

[Petros626]

Hi guys,

I am using the ssd_mobilenet_v2 to train it for my purposes. I noticed that the network only generates axis-aligned bounding boxes. Is it now not possible to use my labelled dataset with bounding boxes that are not axis-aligned?

I assume that TensorFlow has not yet fixed this problem for all models?

[Petros626]

@chunduriv any plans for the future to add this for the whole OD API? Would be great to add this feature to Object Detection, they missed it.