search

nianticlabs / manydepth

[CVPR 2021] Self-supervised depth estimation from short sequences
Other
618 stars 83 forks source link

when i test the image,the result of the depth is black #66

Open 9796l opened 1 year ago

9796l commented 1 year ago

when i use custom dataset (my own dataset) to train, after training ,i used the weight to test,python test_simple.py,but i got the bad result,the mono depth image is black ,when i used source frame to test, the multi depth image is also black.the multi depth result is here 0000000100_disp_multithe costvol_min_multi.jepg is here 0000000100_costvol_min_multi i test many images ,the results are the same.all are black.i guess the weight is wrong,but i used weight19、weight18……，the result are also black,i did not change the code of test_simple.py.But i find when i used the previous weight(long time ago,i used another dataset to train,i got the weight),the result depth image is colourful,it is normal.but the costvol_min_multi.jepg is black. i do not the real reason of this problem.

yunjinli commented 11 months ago

I'm also experiencing the exact same issue. Have you found the reason already? I would be happy to know :)