Closed tianyufang1958 closed 1 year ago
Hi, this might be a naive question. If MAL after training in phase 1 can generate good pseudo labels as instances, why phase 2 is still needed to train the standard instance segmentation model for instance segmentation?
Because stage 1 does not do object detection.
Hi, this might be a naive question. If MAL after training in phase 1 can generate good pseudo labels as instances, why phase 2 is still needed to train the standard instance segmentation model for instance segmentation?