search

mrharicot / monodepth

Unsupervised single image depth prediction with CNNs
Other
2.21k stars 628 forks source link

Strange Evaluatoin Result on Kitti #230

Open aliko70 opened 5 years ago

aliko70 commented 5 years ago

Hi @mrharicot I am getting strangely high error result when I test by the downloaded kitti model. The generated depthmaps both using the given-model as well as my own trained models all look good while their evaluation on kitti_stereo_2015_test_files gives me the same abnormal results...

Testing: python monodepth_main.py --mode test --data_path path_to_kittiDataset/test_data/ \ --filenames_file path_to_kitti_dataset/kitti_stereo_2015_test_files.txt --log_directory $logdir \ --checkpoint_path path_to_models/model_kitti --output_directory $outputdir

Evaluating: python utils/evaluate_kitti.py --split kitti --predicted_disp_path $outputdir/disparities.npy --gt_path utils/data_scene_flow/test_data

-sample of generated depth:

predicted_depth

minmaxdepth

-corresponding gt-depth:

gt_depth

This ends up with very large values in 'thresh' when computing errors.

eval_results

Will really appreciate it if you can get me some hints @dantkz @gosip @Hirico .

Thanks, Ali

Originally posted by @aliko70 in https://github.com/mrharicot/monodepth/issues/199#issuecomment-462136881

NovaMind-Z commented 4 years ago

Hi, I use the pretrained kitti model to test a pciture in KITTI2015 and get the disparity map like this: 000000_10_disp while the ground truth disparity map is as below : 000000_10 Could you please tell me why they are quite different in their pixel value?

ndsclark commented 3 years ago

Hi @aliko70 Please check whether you have changed the format of image containing the ground truth from .png to .jpg. If so, please change it to .png.

deffandchen commented 3 years ago

Hi @aliko70 Please check whether you have changed the format of image containing the ground truth from .png to .jpg. If so, please change it to .png. The ground truth in the original dataset is .png. Do you mean it doesn't need to be changed?

ndsclark commented 3 years ago

Please check whether you have changed the format of image containing the ground truth from .png to .jpg. If so, please change it to .png. The ground truth in the original dataset is .png. Do you mean it doesn't need to be changed?

Yeah，training set need to be changed, but the ground truth needn't to be changed.

deffandchen commented 3 years ago

Please check whether you have changed the format of image containing the ground truth from .png to .jpg. If so, please change it to .png. The ground truth in the original dataset is .png. Do you mean it doesn't need to be changed?

Yeah，training set need to be changed, but the ground truth needn't to be changed.

Thank you very much! I have got the right results.