ImageNet pretrained weights

[lonelysheeep]

I use your source code to reproduce the paper, the result is worse, where <1.25 is only 0.852, is there something wrong with the source code?

[noahzn]

Hi, <1.25=0.852 is abnormal. What PyTorch version do you use? Did you load the pre-trained weights? The source code should work without problems. As far as I know, at least 5 people have reproduced the results.

[lonelysheeep]

1.7.1，I found that the splits in your code are incomplete, I used the splits in mono2 instead of yours, does this matter?

[noahzn]

Yes. This repo only contains the Eigen test files. You need to prepare the train_files.txt and val_files.txt

[lonelysheeep]

So may I ask, if I want to reproduce the code to achieve your effect, what else do I need to do besides downloading the code from github and running it according to the readme instructions? For example in the other question it was mentioned that I need to download the pre-trained model or something else?

[noahzn]

Yes, if you want to use the ImageNet pre-trained weights you need to download the pre-trained model and load it by setting --mypretrain

[lonelysheeep]

Thank you for your answer, I will try again

[lonelysheeep]

Can you provide me with the download path for the ImageNet pre-training model? I've downloaded several but they don't seem to work

[noahzn]

You can find all the download links you need in the README.md file.

[lonelysheeep]

thanks