 |
| GIANTMONSTER gif |
https://www.ibm.com/topics/monte-carlo-simulation
Shiny object AI, lightweight Python
Smartphone to contact lens
Eye reflection/MIT RICE shiny object link https://openaccess.thecvf.com/content/CVPR2023/papers/Tiwary_ORCa_Glossy_Objects_As_Radiance-Field_Cameras_CVPR_2023_paper.pdf
Poker match
Eyeglasses, sunglasses, eyeball reflection utilized zoom
 |
| https://www.contactlensjournal.com/article/S1367-0484(21)00021-7/abstract |
Video m, n
Casino Royale (2006) - Final Poker Scene
DR. NO | "Bond, James Bond" – Sean Connery, Eunice Gayson | James Bond
Las Vegas casino test
SLAM for multiple extended targets using 6G, WIFI, & satellite signal
saline power system
example for ORCa:
slim Python embedded on lens microprocessor example:
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Computing Virtual Cones for case of sphere "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Objective\n",
"Given spherical surface , set of camera rays and camera points, find virtual cones for each camera ray in parallel.\n",
"\n",
"Starting with including required packaged and adding some helper functions"
]
},
{
"cell_type": "code",
"execution_count": 24,
"metadata": {},
"outputs": [],
"source": [
"%matplotlib widget\n",
"import torch\n",
"from torch import autograd\n",
"from functools import partial\n",
"from mpl_toolkits.mplot3d import Axes3D\n",
"import matplotlib.pyplot as plt\n",
"import numpy as np\n",
"import sys\n",
"sys.path.append('.')\n",
"from utils.viz_util import (plot3d_arrow, plot3d_sphere, \n",
" plot3d_point, plot3d_cone)\n",
"\n",
"dot = lambda x,y: (x*y).sum(-1, keepdim=True)\n",
"norm = lambda x: torch.sqrt((x**2).sum(-1, keepdim=True))\n",
"transp = lambda x: x.transpose(-2,-1)\n",
"normalize = lambda x: x/(1e-7+norm(x))\n",
"to_np = lambda x: x[0,0].detach().cpu().numpy()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Given quantities\n",
"\n",
"$N$: Number of primary rays\n",
"\n",
"$P$: Number of intersection points per primary ray\n",
"\n",
"$\\mathbf{o}$: Primary ray origin ($N \\times P \\times 3$)\n",
"\n",
"$\\mathbf{d}$: Primary ray direction ($N\\times P \\times 3$)\n",
"\n",
"$t$: Distance along ray for intersection points($N\\times P \\times 1$) \n",
"\n",
"$r$: Radius of primary cone ($N \\times P \\times 3$)\n",
"\n",
"$\\mathbf{n}$: Surface normals at intersection normals ($N \\times P \\times 3$) \n",
"\n",
"$R$: Radius of curvature at the surface ($N \\times P \\times 3$)\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Defining one primary ray,then repeating it to $N\\times P$ values to test out the performance of this module in ORCa. Overall code should work with $N$ different primary rays"
]
},
{
"cell_type": "code",
"execution_count": 25,
"metadata": {},
"outputs": [],
"source": [
"N = 128\n",
"P = 192\n",
"\n",
"expand_NP = lambda x: x[None,None].expand(N,P,-1)\n",
"\n",
"# Example visualization\n",
"o = expand_NP(torch.cuda.FloatTensor([3.,0.,0.])) # (N,P,3)\n",
"d = expand_NP(normalize(torch.cuda.FloatTensor([-1,0.2,0]))) # (N,P,3)\n",
"r = expand_NP(torch.cuda.FloatTensor([0.05])) # (N,P,1)\n",
"up = expand_NP(torch.cuda.FloatTensor([0.,1.,0]))# Up vector of camera Useful for cone rays"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Here we obtain $t$ value from ray sphere intersection. But in ORCa pipeline, SDFNet would give us t values. Here consider center $c$ and radius of sphere $R$ are known. In ORCa, center won't be needed and radius can be computed from SDF."
]
},
{
"cell_type": "code",
"execution_count": 26,
"metadata": {},
"outputs": [],
"source": [
"def intersect_ray_sphere(o, d, c, R):\n",
" # Sphere given by <x-c,x-c>=R**2\n",
" # Ray given by x = o+td\n",
" # Point of intersection t follows:\n",
" # a2t^2 + a1t + a0 with following values\n",
" a2 = 1.\n",
" a1 = 2*dot(d, o-c)\n",
" a0 = norm(o-c)**2 - R**2\n",
" # Using quadratic formula finding the smaller of the two roots\n",
" t_int = (-a1 - torch.sqrt(a1**2 - 4*a2*a0))/(2*a2)\n",
"\n",
" return o+t_int*d\n",
"\n",
"\n",
"R = expand_NP(torch.cuda.FloatTensor([1.])) # (N,P,1)\n",
"c = expand_NP(torch.cuda.FloatTensor([0., 0., 0.])) # (N,P,3)\n",
"\n",
"p = intersect_ray_sphere(o,d,c,R)\n",
"# t is distance along ray\n",
"t = norm(p-o) # (N,P,1)\n",
"# surface normal points outwards from center\n",
"n = normalize(p-c) # (N, P, 3)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Here we visualize the known quantities. As all the $N \\times P$ values are the same we just visualize the first one. "
]
},
{
"cell_type": "code",
"execution_count": 27,
"metadata": {},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "f3cec8a9cf19491182a91b364c9cf959",
"version_major": 2,
"version_minor": 0
},
"image/png":
MIT License Copyright (c) 2023 Kushagra Tiwary
Project G
Alpha 008
No comments:
Post a Comment