lina-lln / adds-depthnet Goto Github PK

Hello, I have this problem

Can you tell me how to solve it

Hi many thanks your hard work, I tried to train monodepth2 with oxford dataset (day and night) as Table 1. but the result quite bad. Could you show how to train monodepth2 with oxford dataset with your config?

When will you release the code?

Thank you for your github

Now i'm working on monodepth estimation. And i want to use the robotcar dataset after seeing your research to evaluate my reaserch

but i need the original resolution data(1280x960) of the night validation dataset (but you did crop and resize)
(also, in your reaserch, you set the robotcar dataset's 6th split data, so i checked and it didn't fit well)

so if possible, could you help me?

Hi

As shown in your paper, Monodepth2 seems to have better performance than hr-depth,. It makes me a little confused.
Did you use any trick in your training of Monodepth2 (Day)? THX .

def get_depth ( self , folder , frame_index , side , do_flip ):

Hello
Excuse me, the ground truth of the depth here is not implemented, can you please share how to get the ground truth?

Hi,
we are very interested in the CUDA memory during training. Could you provide an approximate value of the GPU memory during training?
Thank you very much!

作者你好，我从你们提供的链接下载到的关于RobotCar验证集深度真值的图片，感觉看上去有很多并不能对应上，或者是有些问题的。比如，下面两个图。

。
我的疑问是，这样的深度真值没问题吗，能起到验证集的作用吗。