search

ialhashim / DenseDepth

High Quality Monocular Depth Estimation via Transfer Learning
https://arxiv.org/abs/1812.11941
GNU General Public License v3.0
1.58k stars 353 forks source link

Test my pictures #42

Closed cww123456789 closed 5 years ago

cww123456789 commented 5 years ago

Hello, I used test.py to test the indoor photos I took, but the output depth map had poor effect, and even the opposite situation occurred. How can I improve his accuracy?Thank you.

ialhashim commented 5 years ago

Could you provide an example or more details?

Keep in mind the generalization performance (quality of estimated depth on images not very similar to training dataset) is still overall not that good.

cww123456789 commented 5 years ago

I tested four images.The two on the left are pictures of plasterwork and jars taken by the camera, and the two on the right are a composite of 3dsmax. Figure_1

ialhashim commented 5 years ago

These images are very different (in terms of lighting, texture, and content) from NYU (the training data). It is expected to obtain such results for most CNN depth estimation methods. If you find a method that performs better on these challenging images let me know, I would interested to compare against them.

cww123456789 commented 5 years ago

Well, thank you very much for your help. Good luck