NVlabs/Deep_Object_Pose

Depth image same as segmentation

Closed this issue · 2 comments

Hi, I am trying to get depth map along with the segmentation but the data generated is not clear in which format it is. I assumed the depth map must have depth value for each pixel and thus it can be used as ground truth depth for the synthetic RGB images and its corresponding segmentation.

But as you can see below both depth and segmentation .exr files have same output.

Please let me know how to achieve the metric depth using nvisii data generation pipeline.

RGB image
00000

Segmentation Image
image

Depth Image
image

please find the attached link to the original files for you reference. google drive

When you converted the images to PNG, the values were cropped, so everything looks black or white after conversion. This is why the EXR format was chosen in the first place: it allows a higher range of pixel values.

If you open the image in an EXR viewer such as https://viewer.openhdr.org/, you can see that the depth image actually has the depth information:

grafik

If you double-click a pixel, you can see the value. For example, near the center of the object, the value is 1.669, which is the depth in meters.

Thank you very much for the explanation @mintar.