BlockGaussian_Quantization/block_fusion.py at main · AzkZzz04/BlockGaussian_Quantization · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
import os
import json
import yaml
import torch
import argparse
import easydict
import numpy as np
import matplotlib.pyplot as plt
from utils.utils import load_gs_ply
from plyfile import PlyData, PlyElement
from scene.gaussian_model import GaussianModel


def calculate_block_transform(block_bbx):
    A, B, C, D = block_bbx[:4, :]
    E, F, G, H = block_bbx[4:, :]
    center = (A + G) / 2    # [3]

    axis_x = (B - A) / np.linalg.norm((B - A))
    axis_z = (D - A) / np.linalg.norm((D - A))
    axis_y = (E - A) / np.linalg.norm((E - A))

    B2W = np.zeros((4, 4), dtype=np.float32)
    B2W[:3, 0], B2W[:3, 1], B2W[:3, 2] = axis_x, axis_y, axis_z
    B2W[:3, 3] = center
    B2W[3, 3] = 1

    W2B = np.linalg.inv(B2W)
    vertices_block = (W2B[:3, :3] @ block_bbx.T).T + W2B[:3, 3]
    x_extent = abs(vertices_block[0, 0])
    y_extent = abs(vertices_block[0, 1])
    z_extent = abs(vertices_block[0, 2])

    return W2B, float(x_extent), float(y_extent), float(z_extent)


def save_gs_ply(plypath, xyz, features_dc, features_rest, opacity, scaling, rotation):
    os.makedirs(os.path.dirname(plypath), exist_ok=True)

    normals = np.zeros_like(xyz)
    f_dc = features_dc.reshape(features_dc.shape[0], -1)
    f_rest = features_rest.reshape(features_rest.shape[0], -1)

    l = ['x', 'y', 'z', 'nx', 'ny', 'nz']
    # All channels except the 3 DC
    for i in range(f_dc.shape[1]):
        l.append('f_dc_{}'.format(i))
    for i in range(f_rest.shape[1]):
        l.append('f_rest_{}'.format(i))
    l.append('opacity')
    for i in range(scaling.shape[1]):
        l.append('scale_{}'.format(i))
    for i in range(rotation.shape[1]):
        l.append('rot_{}'.format(i))
    dtype_full = [(attribute, 'f4') for attribute in l]

    elements = np.empty(xyz.shape[0], dtype=dtype_full)
    attributes = np.concatenate((xyz, normals, f_dc, f_rest, opacity, scaling, rotation), axis=1)
    elements[:] = list(map(tuple, attributes))

    elements = np.empty(xyz.shape[0], dtype=dtype_full)
    attributes = np.concatenate((xyz, normals, f_dc, f_rest, opacity, scaling, rotation), axis=1)
    elements[:] = list(map(tuple, attributes))
    el = PlyElement.describe(elements, 'vertex')
    PlyData([el]).write(plypath)


def merge_blocks(sh_degree, ply_filepaths, merged_plypath):
    xyz_all, features_dc_all, features_extra_all, opacities_all, scales_all, rots_all = [], [], [], [], [], []
    for ply_filepath in ply_filepaths:
        xyz, features_dc, features_extra, opacities, scales, rots = load_gs_ply(sh_degree, ply_filepath)
        xyz_all.append(xyz)
        features_dc_all.append(features_dc)
        features_extra_all.append(features_extra)
        opacities_all.append(opacities)
        scales_all.append(scales)
        rots_all.append(rots)

    xyz_all = np.concatenate(xyz_all, axis=0)
    features_dc_all = np.concatenate(features_dc_all, axis=0)
    features_extra_all = np.concatenate(features_extra_all, axis=0)
    opacities_all = np.concatenate(opacities_all, axis=0)
    scales_all = np.concatenate(scales_all, axis=0)
    rots_all = np.concatenate(rots_all, axis=0)

    print("Num points of merged scene pcd:", xyz_all.shape[0])
    save_gs_ply(merged_plypath, xyz_all, features_dc_all, features_extra_all, opacities_all, scales_all, rots_all)


def plot_rectangle(rectangle: np.ndarray, color=np.random.rand(3)):
    """
    Plot a rectangle given its vertices and fill it with color.

    Parameters:
        rectangle (np.ndarray): An array of shape [4, 2] representing the rectangle's vertices.
        color (str): The fill color of the rectangle.
    """
    # Close the rectangle by appending the first point to the end
    rectangle_closed = np.vstack([rectangle, rectangle[0]])

    # Plot and fill the rectangle
    plt.fill(rectangle_closed[:, 0], rectangle_closed[:, 1], color=color, alpha=0.3)
    plt.plot(rectangle_closed[:, 0], rectangle_closed[:, 1], color=color, alpha=0.3)


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="Block fusion to construct the whole scene.")
    parser.add_argument("--optimized_path", "-o", type=str, default="./output/rubble", help="optimized scene dirpath")
    parser.add_argument("--merge", action="store_true", help="generate scene merged ply file")
    args = parser.parse_args()

    config_filepath = os.path.join(args.optimized_path, "config.yaml")
    with open(config_filepath, "rb") as f:
        cfg = yaml.load(f, Loader=yaml.FullLoader)
    cfg = easydict.EasyDict(cfg)
    cfg.output_dirpath = args.optimized_path

    output_dirpath = args.optimized_path
    blocks_info_jsonpath = os.path.join(cfg.output_dirpath, "blocks_info.json")
    with open(blocks_info_jsonpath, "r") as json_file:
        blocks_info = json.load(json_file)
    sh_degree = cfg.sh_degree
    num_blocks = blocks_info["num_blocks"]

    points_counter = 0
    # prune block outliners with block bounding box
    with torch.no_grad():
        block_gaussian = GaussianModel(sh_degree=sh_degree)
        for block_id in range(0, num_blocks):
            optimized_pcd_filename = sorted(os.listdir(os.path.join(cfg.output_dirpath, "point_cloud", str(block_id))))[-1]
            print(block_id, optimized_pcd_filename)
            optimized_pcdpath = os.path.join(cfg.output_dirpath, "point_cloud", str(block_id), optimized_pcd_filename)
            block_info = blocks_info[str(block_id)]
            bbx = np.array(block_info["bbx"]) # [8, 3]
            W2B, x_extent, y_extent, z_extent = calculate_block_transform(bbx)
            W2B = torch.tensor(W2B, device="cuda")
            block_gaussian.load_ply(optimized_pcdpath)

            xyz_block = (W2B[:3, :3] @ block_gaussian.get_xyz.T).T + W2B[:3, 3]
            mask_position = (xyz_block[:, 0] >= -x_extent) & (xyz_block[:, 0] <= x_extent) & \
                            (xyz_block[:, 2] >= -z_extent) & (xyz_block[:, 2] <= z_extent) & \
                            (xyz_block[:, 1] >= -y_extent*1.2) & (xyz_block[:, 1] <= y_extent*1.2)
            print("block_id:", block_id, ", num pts before postfix:", mask_position.shape[0], ", num pts after postfix:", mask_position.sum().item())
            block_gaussian._xyz = block_gaussian._xyz[mask_position]
            block_gaussian._features_dc = block_gaussian._features_dc[mask_position]
            block_gaussian._features_rest = block_gaussian._features_rest[mask_position]
            block_gaussian._opacity = block_gaussian._opacity[mask_position]
            block_gaussian._scaling = block_gaussian._scaling[mask_position]
            block_gaussian._rotation = block_gaussian._rotation[mask_position]
            block_gaussian.save_ply(os.path.join(cfg.output_dirpath, "point_cloud_postfix", "point_cloud_{}.ply".format(block_id)))
            points_counter += block_gaussian._xyz.shape[0]
    print("Postfix over, scene total points number is: ", points_counter)

    if args.merge:
        # merge all postfix block_pcd to scene_pcd
        merged_plypath = os.path.join(cfg.output_dirpath, "point_cloud_merged.ply")
        ply_filepaths = []
        for block_id in range(0, num_blocks):
            ply_filepaths.append(os.path.join(cfg.output_dirpath, "point_cloud_postfix", "point_cloud_{}.ply".format(block_id)))
        merge_blocks(sh_degree, ply_filepaths, merged_plypath)