I'm trying to detect just dogs in input images. Would a pre-trained network on COCO significantly perform better if it was fine-tuned using COCO again (or another dataset) where all non-dog instances are labelled using a single, unified negative label (assuming the classes are balanced)? Would the fine-tuned network focus less on classification of negative instances, and thus focus more on positive instances?


  • $\begingroup$ In my experience condensing labels will improve performance but that I can't say that is true across the board. Best just to try. $\endgroup$ Aug 1 at 5:31


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.