search

ialhashim / DenseDepth

High Quality Monocular Depth Estimation via Transfer Learning
https://arxiv.org/abs/1812.11941
GNU General Public License v3.0
1.57k stars 357 forks source link

Model output #193

Closed Amitdedhia6 closed 2 years ago

Amitdedhia6 commented 2 years ago

Hi - this is not an issue. But I need more information.

Can anyone explain me what is model output? Is it depth at each pixel in meters? In the following line in utils.py's predict function:

predictions = model.predict(images, batch_size=batch_size)

... I see that the model outputs some numbers which are smaller for far objects and larger for near objects. What does this number mean?

And also what is the purpose of the next line (see below)? np.clip(DepthNorm(predictions, maxDepth=maxDepth), minDepth, maxDepth) / maxDepth

This line effectively computes reciprocal of the model output and limits the values to 1. This helps in generating the depth map as picture. However I am more interested in getting the actual depth value at each pixel. Is the model prediction good enough for getting actual depth value (say in meters)?

laisimiao commented 2 years ago

Hi, @Amitdedhia6 Have you figured out?

Amitdedhia6 commented 2 years ago

Hi @laisimiao, there is an answer available at https://github.com/ialhashim/DenseDepth/issues/163 I tried it, did not work well for me

dsnsabari commented 1 year ago

@Amitdedhia6 , Have you solved the problem?