How to initialization with a random point cloud in my own datasets?

[fanzz1208]

How to initialization with a random point cloud in my own datasets?

[inspirelt]

Two ways: 1. you can refer to

Scaffold-GS/scene/dataset_readers.py

Lines 307 to 319 in da97ef8

	if ply_path is None:
	ply_path = os.path.join(path, "points3d.ply")
	if not os.path.exists(ply_path):
	# Since this data set has no colmap data, we start with random points
	num_pts = 10_000
	print(f"Generating random point cloud ({num_pts})...")
	# We create random points inside the bounds of the synthetic Blender scenes
	xyz = np.random.random((num_pts, 3)) * 2.6 - 1.3
	shs = np.random.random((num_pts, 3)) / 255.0
	pcd = BasicPointCloud(points=xyz, colors=SH2RGB(shs), normals=np.zeros((num_pts, 3)))
	storePly(ply_path, xyz, SH2RGB(shs) * 255)

to give a rand init during loading data;

2. replace

   Scaffold-GS/scene/gaussian_model.py

   Line 237 in da97ef8

   points = pcd.points[::self.ratio]

   with a rand init.

[fanzz1208]

to give a rand init during loading data

So what should I do, just delete points3D.txt from the dataset?
Or, how can I modify this code?
my dataset is "Colmap"
Because my dataset was taken by myself, the point cloud extracted using SFM performs poorly, so I want to test it with a randomly initialized point cloud.

[inspirelt]

To do so, you can replace

Scaffold-GS/scene/dataset_readers.py

Lines 191 to 194 in da97ef8

	try:
	xyz, rgb, _ = read_points3D_binary(bin_path)
	except:
	xyz, rgb, _ = read_points3D_text(txt_path)

with

num_pts = 100_000 
print(f"Generating random point cloud ({num_pts})...")   
xyz = np.random.random((num_pts, 3)) * 2.6 - 1.3  # adjust 2.6 and 1.3 to match your custom scene
rgb = np.random.random((num_pts, 3))

BTW, since your point cloud from Colmap is poor, I'm afraid that your pose is poor as well.

[fanzz1208]

To do so, you can replace

Scaffold-GS/scene/dataset_readers.py

Lines 191 to 194 in da97ef8

	try:
	xyz, rgb, _ = read_points3D_binary(bin_path)
	except:
	xyz, rgb, _ = read_points3D_text(txt_path)

with

num_pts = 100_000 
print(f"Generating random point cloud ({num_pts})...")   
xyz = np.random.random((num_pts, 3)) * 2.6 - 1.3  # adjust 2.6 and 1.3 to match your custom scene
rgb = np.random.random((num_pts, 3))

BTW, since your point cloud from Colmap is poor, I'm afraid that your pose is poor as well.

Thank you for your answer. I still want to give it a try.
The data I am using is poor, so the pose matched with colmap is very poor, and I don't know how to solve it.
Do you have any good suggestions?

[inspirelt]

Can you obtain the gt pose through other method, rather than Comap?