Hi, I find use_coco_data: true in the segmentation task, and this does not seem to be a standard test setting? Could you update the standard results for aligning with the result of deeplabv3 in mmdet, this would be very helpful for us to follow your works~
Deeplabv3 use models pre-trained on MS-COCO and then fine-tune on PASCAL VOC 2012. Instead of doing two stage training, we train models jointly. So, our training recipe is in alignment with the one reported in Deeplab papers.
Hi, I find
use_coco_data: true
in the segmentation task, and this does not seem to be a standard test setting? Could you update the standard results for aligning with the result of deeplabv3 in mmdet, this would be very helpful for us to follow your works~