The pretrained model for testing other dataset?

ialhashim / DenseDepth

High Quality Monocular Depth Estimation via Transfer Learning

https://arxiv.org/abs/1812.11941

GNU General Public License v3.0

1.59k stars 353 forks source link

The pretrained model for testing other dataset? #62

Closed yuzehui1996 closed 5 years ago

yuzehui1996 commented 5 years ago

I use your pretrained model(nyu.h5) to test my own dataset. My dataset might not be the same range as NYU-Depth V2(range 10 meters). My dataset range 0 to 2m. Is it suitable to test my dataset with your pretrained model? If so, how can I change the hyper-parameters(maxDepth and minDepth in utils.predict)?

ialhashim commented 5 years ago

Your milage May vary.. Try it out and see how far off are the estimates in the range of 0 to 10.. The model was trained on a range that intersect with what your data is at. The result however depend on many things. The best result would be to fine tune the model if you have ground truth and train the model some more. Changing the liner scaling of the predictions won't magically make the model more accurate.