1 paper
Seongjun Kang, Ishaan Garg, Vishnu Bharadwaj
Two-stage vision pipelines often place an enhancement network before a task network, on the assumption that a cleaner input produces a better output. We evaluate this in a firefigh…