Not to be confused with multi-class classification, in a multi-label problem some observations can be associated with 2 or more classes. Then, during our last GDD Friday at GoDataDriven I decided to go ahead and start adding the multi-class classification use case. In particular, thanks to the flexibility of the DataFrameIterator class added by this should be possible. This empowerment may come in different ways, such like multi-class classification, multi-label classification, object detection (bounding boxes), segmentation, pose estimation, optical flow, etc.Īfter a small discussion with collaborators of the keras-preprocessing package we decided to start empowering Keras users with some of these use cases through the known ImageDataGenerator class. In order to make AI capable of understanding images in the wild as we do, we must empower AI with all those capabilities.
In order to really "understand" an image there are many factors that play a role, like the amount of objects in the image, their dynamics, the relation between frames, the positions of the objects, etc. Images taken in the wild are extremely complex.
0 kommentar(er)
