search

XinJCheng / CSPN

Convolutional Spatial Propagation Network
495 stars 94 forks source link

why the result of my eval sample looks very bad with load the best_model.pth? #40

Open poseidonv opened 3 years ago

poseidonv commented 3 years ago

when I use cpu run this code,load the best model you give,the result of my eval sample looks very bad.The data I use is nyu_depth_v2_labeled.mat,and I convert the data to image and depth npy file and load it.And the error: Epoch: 0, step: 0, loss=1.3417 MSE=2.7589(2.7589) RMSE=1.6610(1.6610) MAE=1.3417(1.3417) ABS_REL=0.4409(0.4409) DELTA1.02=0.0673(0.0673) DELTA1.05=0.1037(0.1037) DELTA1.10=0.1616(0.1616) DELTA1.25=0.3161(0.3161) DELTA1.25^2=0.5606(0.5606) DELTA1.25^3=0.7404(0.7404)

00000_gt 00000_input 00000_pred

cdrwolfe commented 3 years ago

Same issue it seems as well, not sure why though?

I3aer commented 3 years ago

Same issue. I read his code and two papers (Conf and PAMI). There are some contradictions about Up projection layers in the code and papers. I obtained a similar depth map. Code was not written well. It would be much better to train your code since the author dont reply to the issues.

I3aer commented 3 years ago

I have solved my problem by removing normalization of input rgb image using imagenet mean and std_dev.

nickle-fang commented 2 years ago

@I3aer hi, I have met the same issue. But I can't fix it by removing normalization of input rgb image using imagenet mean and std_dev. Which model did you use?