search

TRI-ML / packnet-sfm

TRI-ML Monocular Depth Estimation Repository
https://tri-ml.github.io/packnet-sfm/
MIT License
1.23k stars 242 forks source link

Can not reproduce self-supervised packnet-sfm( training scratch) from on DDAD #162

Open yaruzz opened 3 years ago

yaruzz commented 3 years ago

Thanks for your great work! I'm quite interested in your excellent work. However, I have some questions. Firstly, I follow the train_ddad.yaml file,using 8 GPUs. But I can not reproduce your results. I can only obtain 0.228 but the number your model can obtain is 0.173. I'm pretty sure that I follow your code and use your environment. I'm so confused that why the results are so different. image Besides, I also check the leaderboard of CVPR21. I notice that the baseline your released is also 0.22 which is more close to our reproduced results. So can we conclude that the baseline of current dataset (you just updated) is 0.22 rather than 0.17? Looking forward to your reply ! Thanks a lot

image

RobinhoodKi commented 3 years ago

I got the same problem. I also can not reproduce the 0.173 abs_rel. I tried several times but the value is between 0.20 and 0.21. Can you guys check the dataset again? Are you sure the model you released used the same training dataset ? @VitorGuizilini

baiyancheng20 commented 2 years ago

I got the same problem. I also can not reproduce the 0.173 abs_rel. I tried several times but the value is between 0.20 and 0.21. Can you guys check the dataset again? Are you sure the model you released used the same training dataset ? @VitorGuizilini

I got the same result. Did you figure out the problem?