would you provide an improved version corresponding to the original paper?
TengFeiHan0 opened this issue · 3 comments
Sorry for late reply, I'm working for a paper with an improved version of packnet along with some new ideas, as is on the way, so code are not avariable currently.
Sorry for late reply, I'm working for a paper with an improved version of packnet along with some new ideas, as is on the way, so code are not avariable currently.
@FangGet which paper do you refer to? do you mean "SEMANTICALLY-GUIDED REPRESENTATION LEARN- ING FOR SELF-SUPERVISED MONOCULAR DEPTH"? it is a paper produced by the same author and published on ICLR2020, which uses a fixed semantic segmentation network to guide our depth prediction module.
Actually, I already reproduced this paper. my results(monodepth2-based not packnet) are a little better than the author's due to the difference of semantically-guided network. if you also are interested in this paper, I'm glad to cooperate with you and make more experiments.
the paper is "PackNet-SfM: 3D Packing for Self-Supervised Monocular Depth Estimation", the same author as your paper. could you give the result you reproduced on KITTI? you can also try to make experiments with packnet-block for better results.
for other things, contact me with email: fangasfrank at gmail dot com