How to render mesh face index?

Question

How to render an image in which pixel stores the index on the corresponding face?

We can render barycentric coordinates image in Parametric (Shader Editor >> Geometry node), but how to access the corresponding face index?

Is it related to ptex_face_id in Attribute?

And there's Object Index, Material Index, but no Face Index in Indexes.

Answer 1

I'll expand on my comment. Here's a partial answer that will give you the polygon index in a shader. Maybe you can use that to do what you want.

First run this script to bake the polygon index into the U coordinate of a UVMap named PolyIndex. The active object should be the mesh you want to bake. You'll have to re-run the script whenever you edit the mesh to get the UVMap up to date though.

import bpy

ob = bpy.context.object
assert ob and ob.type == 'MESH'

if bpy.context.mode != 'OBJECT':
    bpy.ops.object.mode_set(mode='OBJECT')

mesh = ob.data
uvs = [0.0] * (2 * len(mesh.loops))

for poly in mesh.polygons:
    for loopi in poly.loop_indices:
        uvs[2*loopi] = poly.index

if 'PolyIndex' not in mesh.uv_layers:
    mesh.uv_layers.new(name='PolyIndex')

mesh.uv_layers['PolyIndex'].data.foreach_set('uv', uvs)

Then in the shader, you can get the polygon index with this

Answer 2

Base on the answer of scurest, I can get the following result(top: Geometry Parametric, bottom: triangle face index), you can check pixel-level correspondence in tev:

To distinguish from the background, I add an offset to face index: uvs[2 * loopi] = poly.index + 1, and I shuffled the face index to make it easier to visualize(the colors of adjacent faces may be almost the same under 256 different colors):

import cv2
import numpy as np

path_img_idx = "UVMap_TriFaceIndex_01.exr"
img_idx = cv2.imread(path_img_idx, -1)
img_idx = img_idx[:, :, 0].astype(np.int32)
max_idx = np.max(img_idx)

lut = np.arange(1, max_idx + 1)
np.random.shuffle(lut)
lut = np.array([0, *lut]).astype(np.int32) # 0 for background

img_idx = lut[img_idx]  # https://stackoverflow.com/a/67765996/10636347

img_idx = (img_idx / max_idx * 255).astype(np.uint8)
img_idx_vis = cv2.applyColorMap(img_idx, cv2.COLORMAP_JET)
cv2.imwrite(path_img_idx[:-4] + "_vis.png", img_idx_vis)