2:00 PM - 2:20 PM
[1L2-J-11-03] Filtering of Middle Layer Outputs for Object Classification Using a Model Trained Detecting Grasping Positions
Keywords:Deep Learning, Object Picking, Object Classification, Grasping Position
We propose a new method to filter middle layer outputs for object classification using a model trained only detecting grasping positions.
Industrial robots have to detect grasping positions and classify objects at the same time for the picking task, when objects of different shapes are mixed.
One of conventional methods for learning multiple tasks is using middle layer outputs of a deep neural network.
However, simple middle layer outputs can not classify objects, when there are multiple objects in the input image.
As a solution to this problem, we propose a method with filtering of middle layer outputs by backpropagation.
We experiment with the object classification using our method.
The results of our experiment show that our proposed method can classify objects using a model trained only detecting grasping positions, even when the input image has multiple objects.
Industrial robots have to detect grasping positions and classify objects at the same time for the picking task, when objects of different shapes are mixed.
One of conventional methods for learning multiple tasks is using middle layer outputs of a deep neural network.
However, simple middle layer outputs can not classify objects, when there are multiple objects in the input image.
As a solution to this problem, we propose a method with filtering of middle layer outputs by backpropagation.
We experiment with the object classification using our method.
The results of our experiment show that our proposed method can classify objects using a model trained only detecting grasping positions, even when the input image has multiple objects.