Open3d: Image and depth to RGBD, Image has unrecognized bytes_per_channel.

[lelouedec]

Hey,
I am trying to open an image and its depth to a rgbd image using Open3D, but I face the error

[CreateRGBDImageFromColorAndDepth] Unsupported image format.
Color image : 0x0, with 0 channels.
Depth image : 0x0, with 0 channels.
Use numpy.asarray to access buffer data.
terminate called after throwing an instance of 'std::runtime_error'
  what():  Image has unrecognized bytes_per_channel.

I tried to open it using the read function of Open3D or opening it in a numpy array before giving it to open3D but the same error pops up. What I do is the following :

color_raw = img.open("./08_54_36_127/rgb_8UC3.png")
color_raw = np.asarray(color_raw)

depth_raw = img.open("./08_54_36_127/depth_16UC1.png")
depth_raw = np.asarray(depth_raw)

image = Image( (color_raw).astype(np.float32))
depth = Image( (depth_raw).astype(np.float32))

rgbd_image = create_rgbd_image_from_color_and_depth(image,depth)

I can open and see the images using matplotlib before transforming them in the open3D and after.

Any idea ?

Ubuntu 16.04
compiled from source

[qianyizh]

Can you upload the files to dropbox and share them here?

[lelouedec]

https://www.dropbox.com/sh/djpqres4knf30m3/AAA6KsxwvJ_GTaLjn6jLJflaa?dl=0
Here the link to the Data.
I coded the transformation from depth to point cloud myself, and it seems to be working, but would prefer to have the RGBD to pointcloud feature of Open3D.

BTW I changed the "astype" type as uint.

[qianyizh]

@syncle I know you are on vacation, but I need you to help me on this, :)

[syncle]

Hi @lelouedec, you don't have to explicitly change the type. Can you try something like

    color_raw = read_image("../../TestData/RGBD/color/00000.jpg")
    depth_raw = read_image("../../TestData/RGBD/depth/00000.png")
    rgbd_image = create_rgbd_image_from_color_and_depth(
        color_raw, depth_raw);

[lelouedec]

Hi @syncle, thank you for your help.
I tried what you said an got

    depth_raw = read_image(SB_file_depth)
    color_raw = read_image(SB_file_color)
    print("images read")
    rgbd_image = create_rgbd_image_from_color_and_depth(color_raw, depth_raw);

images read
[CreateRGBDImageFromColorAndDepth] Unsupported image format.

[syncle]

Can you check the color image is: 8bit 1channel or 8bit 3channel, and depth image is: 16bit 1channel image? By default Open3D regard value 1000 in depth image as 1000mm, or 1m.

[lelouedec]

The depth image is indeed 16bit and its values are 1000 = 1m.
And the color image is 8 bit 3channels
Images informations

Depth :
{'greyscale': True, 'alpha': False, 'planes': 1, 'bitdepth': 16, 'interlace': 0, 'size': (1280, 720)}

Color :
{'greyscale': False, 'alpha': False, 'planes': 3, 'bitdepth': 8, 'interlace': 0, 'size': (1920, 1080)}

[syncle]

Aha, depth and color images should be aligned and in the same image size. Can you align your color and depth image and try again?

[lelouedec]

Oh ok indeed now the depth aligned and scale to the RGB, it is read correctly.
Thanks a lot for the help.

[Magicyu-2015]

@lelouedec Hi! Could you share your method to aligned the depth and color images in the same image size?

[Sumit564]

@syncle Hi syncle. I am having an issue regarding Volumetric Reconstruction using codes provided in open3d documentation. Error is -- [Open3D ERROR] [ScalableTSDFVolume::Integrate] Unsupported image format.

Depth Dataset Prop - {560 * 400 | Bit Depth- 24 | File size(appx.)- 258KB}
Image Dataset Prop - {560 * 400 | Bit Depth- 24 |File size(appx.)- 330KB}

All images and depth are in .png format and aligned properly.
The code is throwing traceback error.
Can I get a solution at the earliest.

[saber2124]

@syncle Hi, I am trying to capture depth and color image from point cloud and the convert image to point cloud, The method is work when i pass directly open3d image and depth but when i use convert_to_intensity = False i can not use directly open3d image and depth and I have to save image and depth image and then pass the to creat_rgbd_from_color_and_depth which give me low quality point cloud data? any suggestion? thank you very much in advance