I have a few questions about the downstream tasks after implementing the unsupervised pre-train. Can I consider your detection as a downstream task? If so, should it be supervised, unsupervised, or weakly supervised? To my understanding, unsupervised learning is based on the fact that few data are labeled. If so, should fine-tune on downstream tasks include labels?
I have a few questions about the downstream tasks after implementing the unsupervised pre-train. Can I consider your detection as a downstream task? If so, should it be supervised, unsupervised, or weakly supervised? To my understanding, unsupervised learning is based on the fact that few data are labeled. If so, should fine-tune on downstream tasks include labels?