GSMR-CNN: An End-to-End Trainable Architecture for Grasping Target Objects from Multi-Object Scenes | IEEE Conference Publication | IEEE Xplore