jmplaza
commented
5 years ago Hi @rkachach, this labelling capability has been proposed for ObjectDetector, but it is already implemented in DetectionSuite, and this is a better place for it I think. DetectionSuite is a useful tool when working on detection problem in images with neural networks and deep learning.
A new issue about improving its documentation has been created, in order to make it easier to use.
In order to retrain some model (using transfer learning) we normally need to have some dataset or generate one. This normally implies tagging objects in images manually using tools such as label image. Unfortunately, this manual process requires a lot of effort/time since we have to mark manually all the objects within an image and specify their classes.
Object-detector is a powerful tool which already has a good support for different models/frameworks and is able to detect object automatically in video/images. It would be very helpful to extend the tool so it can be used to generate detection datasets automatically (in VOC PASCAL format for example). For this, the tool can detect automatically the objects within a frame and then let the user select which ones are correct and discard the bad ones. This would be very helpful to speedup the process of data labeling.