Sign up to take part
Registered users can ask their own questions, contribute to discussions, and be part of the Community!
Registered users can ask their own questions, contribute to discussions, and be part of the Community!
The object detection algorithm in Dataiku (Faster RCNN) works well at detecting objects labelled in images but the training appears to automatically filter out all images that do not contain an object. This results in major shortcomings in precision. Is it possible to train with a mixture of images that do or do not contain annotated objects? Searching the internet gives me limited results but some threads suggest that this is an option for Faster RCNN: https://github.com/tensorpack/tensorpack/issues/1230