Spaces:
Sleeping
Sleeping
File size: 6,300 Bytes
8e0b903 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 |
import math
import os
from dataclasses import dataclass, field
from typing import List, Union
import numpy as np
import PIL.Image
import torch
import torch.nn.functional as F
import trimesh
from einops import rearrange
from huggingface_hub import hf_hub_download
from omegaconf import OmegaConf
from PIL import Image
import sys
script_dir = os.path.dirname(os.path.abspath(__file__))
imgto3d_path = os.path.join(script_dir, '.')
print(imgto3d_path)
sys.path.append(imgto3d_path)
from .models.isosurface import MarchingCubeHelper
from tsr_utils import (
BaseModule,
ImagePreprocessor,
find_class,
get_spherical_cameras,
scale_tensor,
)
class TSR(BaseModule):
@dataclass
class Config(BaseModule.Config):
cond_image_size: int
image_tokenizer_cls: str
image_tokenizer: dict
tokenizer_cls: str
tokenizer: dict
backbone_cls: str
backbone: dict
post_processor_cls: str
post_processor: dict
decoder_cls: str
decoder: dict
renderer_cls: str
renderer: dict
cfg: Config
@classmethod
def from_pretrained(
cls, config_path: str, weight_path: str
):
cfg = OmegaConf.load(config_path)
OmegaConf.resolve(cfg)
model = cls(cfg)
ckpt = torch.load(weight_path, map_location="cpu")
model.load_state_dict(ckpt)
return model
def configure(self):
self.image_tokenizer = find_class(self.cfg.image_tokenizer_cls)(
self.cfg.image_tokenizer
)
self.tokenizer = find_class(self.cfg.tokenizer_cls)(self.cfg.tokenizer)
self.backbone = find_class(self.cfg.backbone_cls)(self.cfg.backbone)
self.post_processor = find_class(self.cfg.post_processor_cls)(
self.cfg.post_processor
)
self.decoder = find_class(self.cfg.decoder_cls)(self.cfg.decoder)
self.renderer = find_class(self.cfg.renderer_cls)(self.cfg.renderer)
self.image_processor = ImagePreprocessor()
self.isosurface_helper = None
def forward(
self,
image: Union[
PIL.Image.Image,
np.ndarray,
torch.FloatTensor,
List[PIL.Image.Image],
List[np.ndarray],
List[torch.FloatTensor],
],
device: str,
) -> torch.FloatTensor:
rgb_cond = self.image_processor(image, self.cfg.cond_image_size)[:, None].to(
device
)
batch_size = rgb_cond.shape[0]
input_image_tokens: torch.Tensor = self.image_tokenizer(
rearrange(rgb_cond, "B Nv H W C -> B Nv C H W", Nv=1),
)
input_image_tokens = rearrange(
input_image_tokens, "B Nv C Nt -> B (Nv Nt) C", Nv=1
)
tokens: torch.Tensor = self.tokenizer(batch_size)
tokens = self.backbone(
tokens,
encoder_hidden_states=input_image_tokens,
)
scene_codes = self.post_processor(self.tokenizer.detokenize(tokens))
return scene_codes
def render(
self,
scene_codes,
n_views: int,
elevation_deg: float = 0.0,
camera_distance: float = 1.9,
fovy_deg: float = 40.0,
height: int = 256,
width: int = 256,
return_type: str = "pil",
):
rays_o, rays_d = get_spherical_cameras(
n_views, elevation_deg, camera_distance, fovy_deg, height, width
)
rays_o, rays_d = rays_o.to(scene_codes.device), rays_d.to(scene_codes.device)
def process_output(image: torch.FloatTensor):
if return_type == "pt":
return image
elif return_type == "np":
return image.detach().cpu().numpy()
elif return_type == "pil":
return Image.fromarray(
(image.detach().cpu().numpy() * 255.0).astype(np.uint8)
)
else:
raise NotImplementedError
images = []
for scene_code in scene_codes:
images_ = []
for i in range(n_views):
with torch.no_grad():
image = self.renderer(
self.decoder, scene_code, rays_o[i], rays_d[i]
)
images_.append(process_output(image))
images.append(images_)
return images
def set_marching_cubes_resolution(self, resolution: int):
if (
self.isosurface_helper is not None
and self.isosurface_helper.resolution == resolution
):
return
self.isosurface_helper = MarchingCubeHelper(resolution)
def extract_mesh(self, scene_codes, resolution: int = 256, threshold: float = 25.0):
self.set_marching_cubes_resolution(resolution)
meshes = []
for scene_code in scene_codes:
with torch.no_grad():
density = self.renderer.query_triplane(
self.decoder,
scale_tensor(
self.isosurface_helper.grid_vertices.to(scene_codes.device),
self.isosurface_helper.points_range,
(-self.renderer.cfg.radius, self.renderer.cfg.radius),
),
scene_code,
)["density_act"]
v_pos, t_pos_idx = self.isosurface_helper(-(density - threshold))
v_pos = scale_tensor(
v_pos,
self.isosurface_helper.points_range,
(-self.renderer.cfg.radius, self.renderer.cfg.radius),
)
with torch.no_grad():
color = self.renderer.query_triplane(
self.decoder,
v_pos,
scene_code,
)["color"]
mesh = trimesh.Trimesh(
vertices=v_pos.cpu().numpy(),
faces=t_pos_idx.cpu().numpy(),
vertex_colors=color.cpu().numpy(),
)
meshes.append(mesh)
return meshes
|