Spaces:
Build error
Build error
File size: 9,609 Bytes
e206fc8 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 |
import os
import tyro
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from core.options import AllConfigs, Options
from core.gs import GaussianRenderer
import dearpygui.dearpygui as dpg
import kiui
from kiui.cam import OrbitCamera
class GUI:
def __init__(self, opt: Options):
self.opt = opt
self.W = opt.output_size
self.H = opt.output_size
self.cam = OrbitCamera(self.W, self.H, r=opt.cam_radius, fovy=opt.fovy)
self.device = torch.device("cuda")
self.tan_half_fov = np.tan(0.5 * np.deg2rad(opt.fovy))
self.proj_matrix = torch.zeros(4, 4, dtype=torch.float32, device=self.device)
self.proj_matrix[0, 0] = 1 / self.tan_half_fov
self.proj_matrix[1, 1] = 1 / self.tan_half_fov
self.proj_matrix[2, 2] = (opt.zfar + opt.znear) / (opt.zfar - opt.znear)
self.proj_matrix[3, 2] = - (opt.zfar * opt.znear) / (opt.zfar - opt.znear)
self.proj_matrix[2, 3] = 1
self.mode = "image"
self.buffer_image = np.ones((self.W, self.H, 3), dtype=np.float32)
self.need_update = True # update buffer_image
# renderer
self.renderer = GaussianRenderer(opt)
self.gaussain_scale_factor = 1
self.gaussians = self.renderer.load_ply(opt.test_path).to(self.device)
dpg.create_context()
self.register_dpg()
self.test_step()
def __del__(self):
dpg.destroy_context()
@torch.no_grad()
def test_step(self):
# ignore if no need to update
if not self.need_update:
return
starter = torch.cuda.Event(enable_timing=True)
ender = torch.cuda.Event(enable_timing=True)
starter.record()
# should update image
if self.need_update:
# render image
cam_poses = torch.from_numpy(self.cam.pose).unsqueeze(0).to(self.device)
cam_poses[:, :3, 1:3] *= -1 # invert up & forward direction
# cameras needed by gaussian rasterizer
cam_view = torch.inverse(cam_poses).transpose(1, 2) # [V, 4, 4]
cam_view_proj = cam_view @ self.proj_matrix # [V, 4, 4]
cam_pos = - cam_poses[:, :3, 3] # [V, 3]
buffer_image = self.renderer.render(self.gaussians.unsqueeze(0), cam_view.unsqueeze(0), cam_view_proj.unsqueeze(0), cam_pos.unsqueeze(0), scale_modifier=self.gaussain_scale_factor)[self.mode]
buffer_image = buffer_image.squeeze(1) # [B, C, H, W]
if self.mode in ['alpha']:
buffer_image = buffer_image.repeat(1, 3, 1, 1)
buffer_image = F.interpolate(
buffer_image,
size=(self.H, self.W),
mode="bilinear",
align_corners=False,
).squeeze(0)
self.buffer_image = (
buffer_image.permute(1, 2, 0)
.contiguous()
.clamp(0, 1)
.contiguous()
.detach()
.cpu()
.numpy()
)
self.need_update = False
ender.record()
torch.cuda.synchronize()
t = starter.elapsed_time(ender)
dpg.set_value("_log_infer_time", f"{t:.4f}ms ({int(1000/t)} FPS)")
dpg.set_value(
"_texture", self.buffer_image
) # buffer must be contiguous, else seg fault!
def register_dpg(self):
### register texture
with dpg.texture_registry(show=False):
dpg.add_raw_texture(
self.W,
self.H,
self.buffer_image,
format=dpg.mvFormat_Float_rgb,
tag="_texture",
)
### register window
# the rendered image, as the primary window
with dpg.window(
tag="_primary_window",
width=self.W,
height=self.H,
pos=[0, 0],
no_move=True,
no_title_bar=True,
no_scrollbar=True,
):
# add the texture
dpg.add_image("_texture")
# dpg.set_primary_window("_primary_window", True)
# control window
with dpg.window(
label="Control",
tag="_control_window",
width=600,
height=self.H,
pos=[self.W, 0],
no_move=True,
no_title_bar=True,
):
# button theme
with dpg.theme() as theme_button:
with dpg.theme_component(dpg.mvButton):
dpg.add_theme_color(dpg.mvThemeCol_Button, (23, 3, 18))
dpg.add_theme_color(dpg.mvThemeCol_ButtonHovered, (51, 3, 47))
dpg.add_theme_color(dpg.mvThemeCol_ButtonActive, (83, 18, 83))
dpg.add_theme_style(dpg.mvStyleVar_FrameRounding, 5)
dpg.add_theme_style(dpg.mvStyleVar_FramePadding, 3, 3)
# timer stuff
with dpg.group(horizontal=True):
dpg.add_text("Infer time: ")
dpg.add_text("no data", tag="_log_infer_time")
# rendering options
with dpg.collapsing_header(label="Rendering", default_open=True):
# mode combo
def callback_change_mode(sender, app_data):
self.mode = app_data
self.need_update = True
dpg.add_combo(
("image", "alpha"),
label="mode",
default_value=self.mode,
callback=callback_change_mode,
)
# fov slider
def callback_set_fovy(sender, app_data):
self.cam.fovy = np.deg2rad(app_data)
self.need_update = True
dpg.add_slider_int(
label="FoV (vertical)",
min_value=1,
max_value=120,
format="%d deg",
default_value=np.rad2deg(self.cam.fovy),
callback=callback_set_fovy,
)
def callback_set_gaussain_scale(sender, app_data):
self.gaussain_scale_factor = app_data
self.need_update = True
dpg.add_slider_float(
label="gaussain scale",
min_value=0,
max_value=1,
format="%.2f",
default_value=self.gaussain_scale_factor,
callback=callback_set_gaussain_scale,
)
### register camera handler
def callback_camera_drag_rotate(sender, app_data):
if not dpg.is_item_focused("_primary_window"):
return
dx = app_data[1]
dy = app_data[2]
self.cam.orbit(dx, dy)
self.need_update = True
def callback_camera_wheel_scale(sender, app_data):
if not dpg.is_item_focused("_primary_window"):
return
delta = app_data
self.cam.scale(delta)
self.need_update = True
def callback_camera_drag_pan(sender, app_data):
if not dpg.is_item_focused("_primary_window"):
return
dx = app_data[1]
dy = app_data[2]
self.cam.pan(dx, dy)
self.need_update = True
with dpg.handler_registry():
# for camera moving
dpg.add_mouse_drag_handler(
button=dpg.mvMouseButton_Left,
callback=callback_camera_drag_rotate,
)
dpg.add_mouse_wheel_handler(callback=callback_camera_wheel_scale)
dpg.add_mouse_drag_handler(
button=dpg.mvMouseButton_Middle, callback=callback_camera_drag_pan
)
dpg.create_viewport(
title="Gaussian3D",
width=self.W + 600,
height=self.H + (45 if os.name == "nt" else 0),
resizable=False,
)
### global theme
with dpg.theme() as theme_no_padding:
with dpg.theme_component(dpg.mvAll):
# set all padding to 0 to avoid scroll bar
dpg.add_theme_style(
dpg.mvStyleVar_WindowPadding, 0, 0, category=dpg.mvThemeCat_Core
)
dpg.add_theme_style(
dpg.mvStyleVar_FramePadding, 0, 0, category=dpg.mvThemeCat_Core
)
dpg.add_theme_style(
dpg.mvStyleVar_CellPadding, 0, 0, category=dpg.mvThemeCat_Core
)
dpg.bind_item_theme("_primary_window", theme_no_padding)
dpg.setup_dearpygui()
### register a larger font
# get it from: https://github.com/lxgw/LxgwWenKai/releases/download/v1.300/LXGWWenKai-Regular.ttf
if os.path.exists("LXGWWenKai-Regular.ttf"):
with dpg.font_registry():
with dpg.font("LXGWWenKai-Regular.ttf", 18) as default_font:
dpg.bind_font(default_font)
# dpg.show_metrics()
dpg.show_viewport()
def render(self):
while dpg.is_dearpygui_running():
# update texture every frame
self.test_step()
dpg.render_dearpygui_frame()
opt = tyro.cli(AllConfigs)
# load a saved ply and visualize
assert opt.test_path.endswith('.ply'), '--test_path must be a .ply file saved by infer.py'
gui = GUI(opt)
gui.render() |