How do you handle berman's ground-truth
AttackonMuggle opened this issue · comments
Thanks for your reply, but how did you prepare the ground-truth images. Since in evaluate_metrics.py line 41, the gt between 0-255 will all be converted to 0.
Originally posted by @jinjidelinmouren in #10 (comment)
And you set the nan in gt to 0, did you do this for masking these points when calculating the error?
So that was a factor we added because of the pixel range of the depth maps in the Berman's dataset. AFAIR, the depth maps in the dataset are 16 bit images. Therefore, the code in line 41 converts the depth-map to 0-255 range.
And you set the nan in gt to 0, did you do this for masking these points when calculating the error?
Yes, we only consider the non-nan pixels for computing the error and this is done to execute that.
Get it. I also tried using the pretrained model to calculate the error, but couldn't get the results in Table 1. In addition, I also found that using the finetune model to get the depth, there will be white edges at the bottom of the depth map. Is this what affects my calculation of metrics?
I used test.py to get this.
hello I used internet images to test ,the file that i used is test.py to test.I used the pre-trained checkpoint,and i opened the result epoch_0.html ,i didn't find the predicted depth map.i didn't know the reason.when you test,how could you get the depth map?thank you very much.
Hello Honey, I am having the same issue. Please can you point me out what I am doing wrong? I am not using RGBD images. Thanks!
Hello Honey, I am having the same issue. Please can you point me out what I am doing wrong? I am not using RGBD images. Thanks!
Hi, I had the same problem with 3090 before. Then I switched to 2080 and got the right results. For older model frameworks, you might want to try an older graphics card.