OctaveConv.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(96, 192, 32, 64)"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ch_in, ch_out = 128, 256\n",
    "alpha_in, alpha_out = 0.25, 0.25\n",
    "\n",
    "stride = 1\n",
    "kernel_size = 3\n",
    "padding = 1\n",
    "stride = 1\n",
    "\n",
    "hf_ch_in = int(ch_in * (1 - alpha_in))\n",
    "hf_ch_out = int(ch_out * (1 - alpha_out))\n",
    "\n",
    "lf_ch_in = ch_in - hf_ch_in\n",
    "lf_ch_out = ch_out - hf_ch_out\n",
    "\n",
    "hf_ch_in, hf_ch_out, lf_ch_in, lf_ch_out"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "hf_data = torch.zeros((1, hf_ch_in, 32, 32))\n",
    "lf_data = torch.zeros((1, lf_ch_in, 16, 16))\n",
    "\n",
    "if stride == 2:\n",
    "    hf_data = nn.AvgPool2d(2)(hf_data)\n",
    "\n",
    "conv_hh = nn.Conv2d(hf_ch_in, hf_ch_out, kernel_size, padding=padding)\n",
    "conv_hl = nn.Conv2d(hf_ch_in, lf_ch_out, kernel_size, padding=padding)\n",
    "conv_lh = nn.Conv2d(lf_ch_in, hf_ch_out, kernel_size, padding=padding)\n",
    "conv_ll = nn.Conv2d(lf_ch_in, lf_ch_out, kernel_size, padding=padding)\n",
    "\n",
    "hf_conv = conv_hh(hf_data)\n",
    "hf_pool_conv = conv_hl(nn.AvgPool2d(2)(hf_data))\n",
    "lf_conv = conv_lh(lf_data)\n",
    "\n",
    "if stride == 2:\n",
    "    lf_upsample = lf_conv\n",
    "    lf_down = nn.AvgPool2d(2)(lf_data)\n",
    "else:\n",
    "    lf_upsample = F.interpolate(lf_conv, scale_factor=2)\n",
    "    lf_down = lf_data\n",
    "\n",
    "lf_down_conv = conv_ll(lf_down)\n",
    "\n",
    "out_h = hf_conv + lf_upsample\n",
    "out_l = hf_pool_conv + lf_down_conv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(torch.Size([1, 192, 32, 32]), torch.Size([1, 64, 16, 16]))"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "out_h.shape, out_l.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(torch.Size([192, 96, 3, 3]),\n",
       " torch.Size([64, 96, 3, 3]),\n",
       " torch.Size([192, 32, 3, 3]),\n",
       " torch.Size([64, 32, 3, 3]))"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "conv_hh.weight.shape, conv_hl.weight.shape, conv_lh.weight.shape, conv_ll.weight.shape"
   ]
  },
  {
   "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}