{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#| default_exp utils\n",
    "\n",
    "import os\n",
    "from PIL import Image\n",
    "\n",
    "def make_square(image, min_size=512, fill_color=(255, 255, 255, 0)):\n",
    "    '''\n",
    "        Make a square image with signature in the center and black (transparent)\n",
    "        strips on top and bottom. Cycle GAN is trained with images of this format.    \n",
    "    '''\n",
    "    x, y = image.size\n",
    "    size = max(min_size, x, y)\n",
    "    new_im = Image.new('RGBA', (size, size), fill_color)\n",
    "    new_im.paste(image, (int((size - x) / 2), int((size - y) / 2)))\n",
    "    new_im = new_im.resize((512, 512))\n",
    "    return new_im\n",
    "\n",
    "def resize_images(path):\n",
    "    '''\n",
    "        Resize all the images present in path that matches the ips used in cyclegan\n",
    "        training\n",
    "    '''\n",
    "    dirs = os.listdir(path)\n",
    "    for item in dirs:\n",
    "        if os.path.isfile(path+item):\n",
    "            image = Image.open(path+item)\n",
    "            image = make_square(image)\n",
    "            image = image.convert('RGB')\n",
    "            image.save(path+item)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "ename": "ModuleNotFoundError",
     "evalue": "No module named 'torch'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mModuleNotFoundError\u001b[0m                       Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[15], line 2\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[38;5;66;03m#| export\u001b[39;00m\n\u001b[0;32m----> 2\u001b[0m \u001b[38;5;28;01mimport\u001b[39;00m \u001b[38;5;21;01mtorch\u001b[39;00m \n\u001b[1;32m      3\u001b[0m \u001b[38;5;28;01mimport\u001b[39;00m \u001b[38;5;21;01mtorch\u001b[39;00m\u001b[38;5;21;01m.\u001b[39;00m\u001b[38;5;21;01mnn\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m \u001b[38;5;21;01mnn\u001b[39;00m\n\u001b[1;32m      5\u001b[0m \u001b[38;5;28;01mclass\u001b[39;00m \u001b[38;5;21;01mIdentity\u001b[39;00m(nn\u001b[38;5;241m.\u001b[39mModule):\n",
      "\u001b[0;31mModuleNotFoundError\u001b[0m: No module named 'torch'"
     ]
    }
   ],
   "source": [
    "#| export\n",
    "import torch \n",
    "import torch.nn as nn\n",
    "\n",
    "class Identity(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(Identity, self).__init__()\n",
    "        \n",
    "    def forward(self, x):\n",
    "        return x\n",
    "    \n",
    "\n",
    "class SiameseNetwork(nn.Module):\n",
    "    def __init__(self, model, embedding_size=2000):\n",
    "        super(SiameseNetwork, self).__init__()\n",
    "\n",
    "        self.backbone = model\n",
    "        self.rgb_grayscale = nn.Conv2d(1,3,kernel_size=3,stride=1,padding=1)\n",
    "        self.a = nn.Sigmoid()\n",
    "        self.fc1 = nn.Linear(in_features=1280, out_features=embedding_size)        \n",
    "        self.fc2 = nn.Linear(in_features=embedding_size,out_features=1)    \n",
    "        self.relu = nn.ReLU()\n",
    "        self.backbone.classifier = Identity()\n",
    "\n",
    "    def forward_once(self, x):\n",
    "        \n",
    "        x = self.rgb_grayscale(x)\n",
    "        resnet_output = self.backbone(x)\n",
    "        \n",
    "        return resnet_output\n",
    "        \n",
    "    def forward(self, x1,x2):\n",
    "        \n",
    "        x1 = self.forward_once(x1)\n",
    "        x2 = self.forward_once(x2)\n",
    "        out1, out2 = self.relu(self.fc1(x1)), self.relu(self.fc1(x2))\n",
    "        dis = torch.abs((out1 - out2))\n",
    "        pred = self.a(self.fc2(dis)).squeeze()\n",
    "        return pred"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "255\n",
      "0\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import cv2\n",
    "import numpy as np\n",
    "p = \"/Users/markus/code/explore/signify/results/media/documents/ajy01c00.png\"\n",
    "pnew = \"/Users/markus/code/explore/signify/results/media/documents/3.png\"\n",
    "\n",
    "img = 255 - cv2.imread(p) \n",
    "print(img.max())\n",
    "print(img.min())\n",
    "cv2.imwrite(pnew, img)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#| hide\n",
    "import nbdev; nbdev.nbdev_export()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.10.8 ('base')",
   "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": "3.10.8"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "a2b499d73ef2075ded5123ca15bce54816687eb05fa98ad819942f2bd35191e9"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}