Test Representation
Sue-Tang-Up opened this issue · comments
I'm going to put the refinemask to the task of building extraction, but there exists a problem when presenting the test results. Obviously, many predicted boxes are overlay so I debug the code and find that the a nms operation is missed to process the bounding boxes. Besides, masks are not given in some masks and I still cannot find a reason to explain.
I'm looking forward to your responses since I'm not sure whether some details of the code are missed by myself.
Sorry，I have found the nms part and use it. But I still can not solve the problem [sad]
I think it can be ascribed to the quality of the dataset!,I change the dataset and then the results are better!
I think it can be ascribed to the quality of the dataset!,I change the dataset and then the results are better!
Sorry for replying so late.
Yeah, the better the dataset quality, the better the segmentation results.
In our paper, the improvements of RefineMask w.r.t Mask R-CNN across three datasets: Cityscapes > LVIS > COCO.