in TensorFlow object detection API, is there any possibility to get other shape of recognized object than rectangle?
For example when I'm teaching my computer to recognize a forest, mostly it should be more useful to get some kind of polygon instead of rectangle.
Other thing is for example river, but polygon seems much easier than line.
The simple answer is no. This framework based on Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks and SSD: Single Shot MultiBox Detector which train additional output of CNN for bounding box regression. Can you create CNN with some structure for detect polygon instead of bounding box? May be, but whis will another network.