Create a segmentation picture with each object class rendered in different color ?

Question

I am using the blender to create a visual parking scene model, by using the python script the model could produces a RGB picture of the parking scene as well as the depth image, both are showing below.

I am just wondering would it possible to also create a segmentation image which separates each object class with a color, for example all the car object are in red, all the ground are in blue, all the building are in grey, like the examples

Answer 1

Give each object a unique 'pass index'. You can use the compositor, the Object Info node, and some math nodes to convert it to a colour.

This happens when you divide the object index by the maximum index you're using (to get it to a 0-1 value), and apply it to a diffuse material:

By using a Colour Ramp you can take that 0-1 value and map it to discrete colours:

By swapping out the Diffuse shader with an Emission shader, you just get the straight colour values:

This is the final material:

This is the blend file I used:

See the manual for more info.

Answer 2

A python package bpycv can render annotations for instance segmentation and panoptic segmentation.

First, follow the instruction to install bpycv in Blender.

Here is a simple code to show how to get segmentation annotations. Copy-paste this code to Text Editor and click Run Script button(or Alt+P)

import cv2
import bpy
import bpycv
import random

# remove all MESH objects
[bpy.data.objects.remove(obj) for obj in bpy.data.objects if obj.type == "MESH"]

for inst_id in range(1, 20):
    # create cube and sphere as instance at random location
    location = [random.random() * 4 - 2 for _ in range(3)]
    if inst_id % 2:
        bpy.ops.mesh.primitive_cube_add(size=0.5, location=location)
    else:
        bpy.ops.mesh.primitive_uv_sphere_add(radius=0.5, location=location)
    obj = bpy.context.active_object
    # set each instance a unique inst_id, which is used to generate instance annotation.
    obj["inst_id"] = inst_id

# render image, instance annoatation and depth in one line code
result = bpycv.render_data()

# save result
cv2.imwrite("demo-rgb.jpg", result["image"])
cv2.imwrite("demo-inst.png", result["inst"])
# normalizing depth
cv2.imwrite("demo-depth.png", result["depth"] / result["depth"].max() * 255)

# visualization inst|rgb|depth for human
cv2.imwrite("demo-vis(inst|rgb|depth).jpg", result.vis())

The visualization result of instance segmentation, RGB image and depth

More detial: bpycv