Will low-quality input have a significant impact on the L4GM's results? #4
Description
Greets, thanks the advanced work! In the appendix of paper (section J. More discussions). The paper said "Since ImageDream is a probabilistic generative model, even the same image input could produce very different multi-view generation. Therefore, we believe that such image input could produce very different multi-view generation. Therefore, we believe that such temporally inconsistent multi-view videos are not ideal input to the reconstruction model." I'm curious that did you conduct some experiments in your paper to explore the impact of low-quality multi-view videos (such as the temporally inconsistent multi-view videos) on L4GM?