search

nianticlabs / manydepth

[CVPR 2021] Self-supervised depth estimation from short sequences
Other
607 stars 84 forks source link

On the pre training model of cityscapes #74

Open WangXuCh opened 10 months ago

WangXuCh commented 10 months ago

Can you provide the weights trained on the cityscapes dataset? I downloaded the compressed package you provided, but it does not contain mono weights (Mono_depth, Mono_encoder).

WangXuCh commented 10 months ago

I used the mono weight from the KITTI you provided on the cityscapes, and the problem has been resolved.

May I ask if you could provide the logs of your training on KITTI and cityscapes?

WangXuCh commented 10 months ago

@JamieWatson683 .Hi, I used the pre training weights you provided to replicate the indicators on the KITTI dataset, and the results are as follows. It looks very different from the results in the paper. May I ask what I did wrong? 1694654122397

hutingz commented 4 months ago

Can you provide the weights trained on the cityscapes dataset? I downloaded the compressed package you provided, but it does not contain mono weights (Mono_depth, Mono_encoder).

image

Hello, can you conduct a test on CityScapes? I followed README.md to perform the CityScapes data preparation work, but I don't understand what the data after processing with SfMLearn master. data. prepare_tain_data.py is. It generates. jpg and camera parameters for training, testing, and evaluating all images. Additionally, how was the GT required for testing obtained and how was it processed into. npy (the GT used in evaluate-depth. py Cityscapes is in. npy format, but it is not specified how it was obtained)