grid_sample.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import torch\n",
    "import matplotlib.pyplot as plt\n",
    "import torch.nn.functional as F"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "N = 13\n",
    "C = 3\n",
    "Hin = 512\n",
    "Win = 256\n",
    "Hout = 255\n",
    "Wout = 128\n",
    "# input = np.arange(0, N * C * Hin * Win, dtype=np.float32).reshape(N, C, Win, Hin).transpose([0, 1, 3, 2])\n",
    "input = np.arange(0, N * C * Hin * Win, dtype=np.float32).reshape(N, C, Hin, Win)\n",
    "input.shape, input[0, 0, :, 0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "Sxx = input[0, 0, :, :]\n",
    "plt.pcolormesh(np.arange(Win), np.arange(Hin), Sxx)\n",
    "plt.ylabel('Frequency [samples]')\n",
    "plt.xlabel('Time [samples]')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "x = np.linspace(-1, 1, Wout)\n",
    "y = np.linspace(-1, 1, Hout)\n",
    "pow_y = (np.power(21, (y + 1) / 2) - 11) / 10\n",
    "log_y = 2 * np.log(10 * y + 11) / np.log(21)  - 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "plt.plot(y, pow_y, \"b\", y, log_y , \"g\")\n",
    "plt.show()\n",
    "plt.plot(x, x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# xx, yy = np.meshgrid(np.arange(0, 512), np.arange(0, 256), indexing='xy')\n",
    "xx, yy = np.meshgrid(x, pow_y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "grid = np.empty(shape=(N, Hout, Wout, 2), dtype=np.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "grid[0, :, :, 0] = xx\n",
    "grid[0, :, :, 1] = yy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "grid[0, :, 0, 1] # changes in H axis\n",
    "# grid[0, 0, :, 0] # changes in W axis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "out = F.grid_sample(torch.as_tensor(input), torch.as_tensor(grid), mode='nearest')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "out[0, 0, :, 0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "Sxx = out[0, 0, :, :]\n",
    "plt.pcolormesh(np.arange(Wout), np.arange(Hout), Sxx)\n",
    "plt.ylabel('Frequency [samples]')\n",
    "plt.xlabel('Time [samples]')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "out = F.grid_sample(torch.as_tensor(input), torch.as_tensor(grid), mode='bilinear')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(out[0, 0, :, 2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# normalize ix, iy from [-1, 1] to [0, IH-1] & [0, IW-1]\n",
    "\n",
    "Nval = 0\n",
    "Cval = 0\n",
    "Hval = 10\n",
    "Wval = 2\n",
    "print(grid[0, Hval, Wval, 1])\n",
    "print(grid[0, Hval, Wval, 0])\n",
    "\n",
    "iy = grid[0, Hval, Wval, 1]\n",
    "ix = grid[0, Hval, Wval, 0]\n",
    "ix = ((ix + 1) / 2) * (Win-1);\n",
    "iy = ((iy + 1) / 2) * (Hin-1);\n",
    "\n",
    "print(iy)\n",
    "print(ix)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get NE, NW, SE, SW pixel values from (x, y)\n",
    "ix_nw = int(np.floor(ix))\n",
    "iy_nw = int(np.floor(iy))\n",
    "ix_ne = int(ix_nw + 1)\n",
    "iy_ne = int(iy_nw)\n",
    "ix_sw = int(ix_nw)\n",
    "iy_sw = int(iy_nw + 1)\n",
    "ix_se = int(ix_nw + 1)\n",
    "iy_se = int(iy_nw + 1)\n",
    "\n",
    "print((ix_nw, iy_nw), (ix_ne, iy_ne), (ix_sw, iy_sw), (ix_se, iy_se))\n",
    "\n",
    "pts = [(ix_nw, iy_nw), (ix_ne, iy_ne), (ix_sw, iy_sw), (ix_se, iy_se)]\n",
    "plt.scatter(*zip(*pts))\n",
    "plt.scatter([ix], [iy], c=\"r\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get surfaces to each neighbor:\n",
    "nw = (ix_se - ix)    * (iy_se - iy);\n",
    "ne = (ix    - ix_sw) * (iy_sw - iy);\n",
    "sw = (ix_ne - ix)    * (iy    - iy_ne);\n",
    "se = (ix    - ix_nw) * (iy    - iy_nw);\n",
    "\n",
    "print(nw, ne, sw, se)\n",
    "print(np.sum([nw, ne, sw, se]))\n",
    "\n",
    "# clip coordinates when border mode else pad with zeros"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# calculate bilinear weighted pixel value and set output pixel\n",
    "\n",
    "# this works for each C in the input independently\n",
    "nw_val = input[Nval, Cval, iy_nw, ix_nw]\n",
    "ne_val = input[Nval, Cval, iy_ne, ix_ne]\n",
    "sw_val = input[Nval, Cval, iy_sw, ix_sw]\n",
    "se_val = input[Nval, Cval, iy_se, ix_se]\n",
    "print(\"Neighbour vals: \", nw_val, ne_val, sw_val, se_val)\n",
    "out_val = nw_val * nw + ne_val * ne + sw_val * sw + se_val * se;\n",
    "print(out_val)\n",
    "print(out[Nval, Cval, Hval, Wval])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "Sxx = out[0, 0, :, :]\n",
    "plt.pcolormesh(np.arange(Wout), np.arange(Hout), Sxx)\n",
    "plt.ylabel('Frequency [samples]')\n",
    "plt.xlabel('Time [samples]')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}