Is object detection/segmentation supported, i.e. is there any annotation format for such use cases?
There is no pre-trained model which is suited for object detection/segmentation in Deep Learning Studio. So you will have to make the model for such a task yourself.
Regarding annotation format, you could pass co-ordinates of the bounding box and class as (x1;y1;x2;y2;class) and choose Array as the data type. This has a limitation that only one bounding box per image can be done this way.
In future releases, we plan to include pre-trained models for object detection(RetinaNet) and support for outputting multiple bounding boxes.
In the meantime, you can checkout https://github.com/fizyr/keras-retinanet