Depth measurement

[hello7623]

I have evaluated the depth model you gave ，why is it different? What might be wrong?
My device is 1080ti and the model for the evaluation is k_depth.tar, and the code used is your evaluation code

[hello7623]

And I found that I could not reproduce the results of your depth model，I trained from the beginning according to the code you gave me without any changes. What might be wrong?

[JiawangBian]

1. The first problem may be little device difference, as you can see that the results are very similar. You can report your reproduced results in future paper for fair comparison.
2. I guess that it is related to train/val partion, but I tried several partions that give similar results. How is about your reproduced results? Is it far from the reported?

[hello7623]

I am trying again and see what happens
my last result is

[hello7623]

1. The first problem may be little device difference, as you can see that the results are very similar. You can report your reproduced results in future paper for fair comparison.
2. I guess that it is related to train/val partion, but I tried several partions that give similar results. How is about your reproduced results? Is it far from the reported?

I know problem about out on this division, with the use of set, so caused some randomness, which eventually led to the final result of the randomness of different, but the repetition of your thesis is vital tomy research, because I need him as the pretrained model, so can you give me your train, val list, or do you have any other good idea?

[Rashfu]

Hi, @JiawangBian , @hello7623 :
I also meet the problem that I could not reproduce the results of depth model in paper. The result of my reproduced model is far from the reported as follow: (I may use the different kitti train/val set)

abs_rel	sq_rel	rmse	rmse_log	a1	a2	a3
0.284	0.273	9.412	0.404	0.481	0.756	0.892

I will use the kitti256 you release and try it again