Handbag_Brand_Detection.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "ename": "ImportError",
     "evalue": "No module named 'model'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mImportError\u001b[0m                               Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-61-a1bc77078296>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     13\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mwarnings\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     14\u001b[0m \u001b[0mwarnings\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfilterwarnings\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"ignore\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 15\u001b[0;31m \u001b[0;32mfrom\u001b[0m \u001b[0mmodel\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mget_model\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mImportError\u001b[0m: No module named 'model'"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "from torch.autograd import Variable\n",
    "import numpy as np\n",
    "from PIL import Image\n",
    "import os, glob\n",
    "from tqdm import tqdm\n",
    "import torchvision\n",
    "from torchvision import datasets, transforms\n",
    "from matplotlib.pyplot import imshow\n",
    "from torch.utils.data.sampler import SubsetRandomSampler\n",
    "import torch.nn.functional as F\n",
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "image_size = 224\n",
    "batch_size = 16\n",
    "num_workers = 4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'/Users/yuxiong/Desktop/Capstone/Fashion_Apparel_Detection'"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pwd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],\n",
    "                                  std=[0.229, 0.224, 0.225])\n",
    "transform_train = transforms.Compose([transforms.Resize((image_size, image_size), interpolation=Image.ANTIALIAS), \\\n",
    "                      transforms.RandomHorizontalFlip(),\n",
    "                      transforms.ToTensor(), \n",
    "                      normalize])\n",
    "transform_test = transforms.Compose([transforms.Resize((image_size, image_size), interpolation=Image.ANTIALIAS), \\\n",
    "                      transforms.ToTensor(), \n",
    "                      normalize])\n",
    "\n",
    "# data_train = datasets.ImageFolder('/scratch/rw2268/Handbag_brand/data/train', transform=transform_train)\n",
    "# train_loader = torch.utils.data.DataLoader(data_train, batch_size=batch_size, \n",
    "#                                                          shuffle=True, \n",
    "#                                                          num_workers=num_workers)\n",
    "\n",
    "# data_val = datasets.ImageFolder('/scratch/rw2268/Handbag_brand/data/val', transform=transform_test)\n",
    "# val_loader = torch.utils.data.DataLoader(data_val, batch_size=batch_size, \n",
    "#                                                      shuffle=False, \n",
    "#                                                      num_workers=num_workers)\n",
    "\n",
    "data_test = datasets.ImageFolder(\"/Users/yuxiong/Desktop/Capstone/Fashion_Apparel_Detection/CNN-Classifier-Pytorch/data/test\", transform=transform_test)\n",
    "test_loader = torch.utils.data.DataLoader(data_test, batch_size=batch_size, \n",
    "                                                       shuffle=False, \n",
    "                                                       num_workers=num_workers)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "def solver(data_loader, model):\n",
    "    model.eval()\n",
    "   \n",
    "    count_test = 0\n",
    "    test_out = []\n",
    "    \n",
    "    for batch_idx, (data,target) in enumerate(data_loader):\n",
    "        data = Variable(to_cuda(data), volatile=True)\n",
    "        output = model(data)\n",
    "        _, arg_max_out = torch.max(output.data.cpu(), 1)\n",
    "        \n",
    "        \n",
    "        for oo in arg_max_out:\n",
    "            test_out.append('%s,%d\\n'%(str(count_test).zfill(4), oo))\n",
    "            count_test+=1\n",
    "\n",
    "    print(\"test_out: \",test_out)\n",
    "\n",
    "    return test_out\n",
    "\n",
    "def test(val_loader, test_loader, model):\n",
    "    #assert start_epoch>0, \"you must first TRAIN\"\n",
    "    #solver(config.model, val_loader, model,  mode='val')\n",
    "    solver(test_loader, model)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = torchvision.models.resnet18(pretrained=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "ename": "RuntimeError",
     "evalue": "Error(s) in loading state_dict for ResNet:\n\tUnexpected key(s) in state_dict: \"classifier.0.weight\", \"classifier.0.bias\". ",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mRuntimeError\u001b[0m                              Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-55-d70f27f92fcd>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload_state_dict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"/Users/yuxiong/Desktop/Capstone/Fashion_Apparel_Detection/yolov3/0012.pth\"\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmap_location\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mlambda\u001b[0m \u001b[0mstorage\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mloc\u001b[0m\u001b[0;34m:\u001b[0m \u001b[0mstorage\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      2\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/anaconda3/lib/python3.5/site-packages/torch/nn/modules/module.py\u001b[0m in \u001b[0;36mload_state_dict\u001b[0;34m(self, state_dict, strict)\u001b[0m\n\u001b[1;32m    717\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0merror_msgs\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m>\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    718\u001b[0m             raise RuntimeError('Error(s) in loading state_dict for {}:\\n\\t{}'.format(\n\u001b[0;32m--> 719\u001b[0;31m                                self.__class__.__name__, \"\\n\\t\".join(error_msgs)))\n\u001b[0m\u001b[1;32m    720\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    721\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0mparameters\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mRuntimeError\u001b[0m: Error(s) in loading state_dict for ResNet:\n\tUnexpected key(s) in state_dict: \"classifier.0.weight\", \"classifier.0.bias\". "
     ]
    }
   ],
   "source": [
    "model.load_state_dict(torch.load(\"/Users/yuxiong/Desktop/Capstone/Fashion_Apparel_Detection/yolov3/0012.pth\", map_location=lambda storage, loc: storage))\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'train_loader' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-42-5d8e1b8a12f2>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mnum_classes\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtrain_loader\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdataset\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mclasses\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      2\u001b[0m \u001b[0mnum_classes\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0mtrain_loader\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdataset\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mclasses\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'train_loader' is not defined"
     ]
    }
   ],
   "source": [
    "num_classes=len(train_loader.dataset.classes)\n",
    "num_classes\n",
    "train_loader.dataset.classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x2b2c8f363780>"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sp_img = data_train.__getitem__(345)[0].numpy()\n",
    "# img = Image.fromarray(sp_img, 'RGB')\n",
    "sp_img = sp_img.transpose([1,2,0])\n",
    "imshow(sp_img)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load a pre-trained CNN and Train model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')\n",
    "# device = 'cpu'\n",
    "num_epochs = 10\n",
    "num_epochs_decay = 60\n",
    "learning_rate = 0.001\n",
    "stop_training = 3\n",
    "def update_lr(lr, optimizer):\n",
    "    for param_group in optimizer.param_groups:\n",
    "        param_group['lr'] = lr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def remove_layer(model, num_classes):\n",
    "    modules = model.modules()\n",
    "    for m in modules:\n",
    "        if isinstance(m, nn.Linear) and m.weight.data.size()[0]==1000:\n",
    "            w1 = m.weight.data[:num_classes]\n",
    "            b1 = m.bias.data[:num_classes]\n",
    "    try:\n",
    "        if type(model.classifier)==nn.Sequential:\n",
    "            mod = list(model.classifier) #Alexnet, VGG\n",
    "        else:\n",
    "            mod = [model.classifier] #DenseNet\n",
    "\n",
    "    except: \n",
    "        mod = [model.fc] #ResNet\n",
    "\n",
    "    weight = mod[-1].weight.size(1)\n",
    "    mod.pop()\n",
    "    mod.append(torch.nn.Linear(weight,num_classes))\n",
    "    new_classifier = torch.nn.Sequential(*mod)\n",
    "    model.classifier = new_classifier\n",
    "    modules = model.modules()\n",
    "    flag = False\n",
    "    for m in modules:\n",
    "        if isinstance(m, nn.Linear) and m.weight.data.size()[0]==num_classes:\n",
    "            m.weight.data = w1\n",
    "            m.bias.data = b1  \n",
    "            flag = True\n",
    "    assert flag"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "model = torchvision.models.resnet18(pretrained=True)\n",
    "remove_layer(model, num_classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def save_model(model, name, epoch):\n",
    "    dir_ = os.path.join('/scratch/rw2268/Handbag_brand/snapshot', name, '%s.pth'%(str(epoch).zfill(4)))\n",
    "    create_folder(os.path.dirname(dir_))\n",
    "    torch.save(model.state_dict(), dir_)\n",
    "    print('!!Saving model: {}!'.format(dir_))\n",
    "def create_folder(folder):\n",
    "    if not os.path.isdir(folder):\n",
    "        os.makedirs(folder)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'learning_rate' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-47-514780325f5c>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     46\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     47\u001b[0m \u001b[0;31m#==========================================================================#\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 48\u001b[0;31m \u001b[0;32mdef\u001b[0m \u001b[0mtrain\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mtrain_loader\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mval_loader\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlearning_rate\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mlearning_rate\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     49\u001b[0m     \u001b[0mval_before\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     50\u001b[0m     \u001b[0moptimizer\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0moptim\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mAdam\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mparameters\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlearning_rate\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;36m0.5\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m0.999\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'learning_rate' is not defined"
     ]
    }
   ],
   "source": [
    "def solver(name, data_loader, model, epoch, optimizer=None, mode='train'):\n",
    "    model = model.to(device)\n",
    "    if optimizer is None: model.eval()\n",
    "    else: model.train()\n",
    "    loss_cum = []\n",
    "    Acc = 0\n",
    "    count_test = 0\n",
    "    test_out = []\n",
    "    Loss = nn.CrossEntropyLoss()\n",
    "    for batch_idx, (data,target) in tqdm(enumerate(data_loader), \n",
    "                total=len(data_loader), desc=\"!{} -> [{}] Epoch: {}\".format(name.upper(), mode.upper(),epoch)):\n",
    "        volatile = True if optimizer is None else False\n",
    "        data = Variable(data.to(device), volatile=volatile)\n",
    "        target = Variable(target.to(device), volatile=volatile)\n",
    "       \n",
    "        output = model(data)\n",
    "        loss = Loss(output,target)   \n",
    "        \n",
    "        if optimizer is not None:\n",
    "            optimizer.zero_grad()\n",
    "            loss.backward()\n",
    "            optimizer.step()\n",
    "       \n",
    "\n",
    "        loss_cum.append(loss.data.cpu()[0])\n",
    "        _, arg_max_out = torch.max(output.data.cpu(), 1)\n",
    "      \n",
    "        if mode=='test':\n",
    "            for oo in arg_max_out:\n",
    "                test_out.append('%s,%d\\n'%(str(count_test).zfill(4), oo))\n",
    "                count_test+=1\n",
    "\n",
    "        Acc += arg_max_out.long().eq(target.data.cpu().long()).sum()\n",
    "#         if batch_idx % 20 == 0:\n",
    "#             print('Epoch:{}, batch:{}, Acc:{}'.format(epoch+1,batch_idx+1,Acc*100/)\n",
    "    ACC = float(Acc*100)/len(data_loader.dataset)\n",
    "    LOSS = np.array(loss_cum).mean()\n",
    "    if mode=='test':\n",
    "        f=open(os.path.join('snapshot', name, 'test.txt'),'w')\n",
    "        for line in test_out: f.writelines(line)\n",
    "        f.close()\n",
    "    else:\n",
    "        print(\"LOSS %s: %0.3f || ACC %s: %0.2f\"%(mode.upper(), LOSS, mode.upper(), ACC))\n",
    "    \n",
    "    return ACC\n",
    "\n",
    "#==========================================================================#\n",
    "def train(name,train_loader, val_loader, model, learning_rate=learning_rate):\n",
    "    val_before = 0\n",
    "    optimizer = torch.optim.Adam(model.parameters(), learning_rate, [0.5, 0.999])\n",
    "    model.train()\n",
    "    for epoch in range(num_epochs):\n",
    "        solver(name, train_loader, model, epoch, optimizer=optimizer, mode='train')\n",
    "        val_acc = solver(name, val_loader, model, epoch, mode='val')\n",
    "        if val_acc>val_before:\n",
    "            save_model(model, name, epoch+1)\n",
    "            val_before=val_acc\n",
    "            flag_stop=0\n",
    "        else:\n",
    "            flag_stop+=1\n",
    "\n",
    "        if flag_stop==stop_training: \n",
    "            return\n",
    "\n",
    "    # Decay learning rate\n",
    "        if (epoch+1) > (num_epochs - num_epochs_decay):\n",
    "            learning_rate -= (learning_rate / float(num_epochs_decay))\n",
    "            update_lr(learning_rate, optimizer)\n",
    "            print ('Decay learning rate to: {}.'.format(learning_rate))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "!RESNET18 -> [TRAIN] Epoch: 0: 100%|██████████| 314/314 [03:39<00:00,  1.43it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS TRAIN: 1.877 || ACC TRAIN: 44.56\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [VAL] Epoch: 0: 100%|██████████| 50/50 [00:33<00:00,  1.48it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS VAL: 1.989 || ACC VAL: 44.50\n",
      "!!Saving model: /scratch/rw2268/Handbag_brand/snapshot/ResNet18/0001.pth!\n",
      "Decay learning rate to: 0.0009833333333333335.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "!RESNET18 -> [TRAIN] Epoch: 1: 100%|██████████| 314/314 [03:24<00:00,  1.54it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS TRAIN: 1.223 || ACC TRAIN: 63.02\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [VAL] Epoch: 1: 100%|██████████| 50/50 [00:30<00:00,  1.63it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS VAL: 1.371 || ACC VAL: 58.03\n",
      "!!Saving model: /scratch/rw2268/Handbag_brand/snapshot/ResNet18/0002.pth!\n",
      "Decay learning rate to: 0.0009669444444444445.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [TRAIN] Epoch: 2: 100%|██████████| 314/314 [03:24<00:00,  1.31it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS TRAIN: 1.007 || ACC TRAIN: 69.66\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [VAL] Epoch: 2: 100%|██████████| 50/50 [00:30<00:00,  2.49it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS VAL: 1.390 || ACC VAL: 57.14\n",
      "Decay learning rate to: 0.0009508287037037038.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [TRAIN] Epoch: 3: 100%|██████████| 314/314 [03:24<00:00,  1.54it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS TRAIN: 0.851 || ACC TRAIN: 74.11\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [VAL] Epoch: 3: 100%|██████████| 50/50 [00:30<00:00,  1.63it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS VAL: 1.397 || ACC VAL: 58.91\n",
      "!!Saving model: /scratch/rw2268/Handbag_brand/snapshot/ResNet18/0004.pth!\n",
      "Decay learning rate to: 0.0009349815586419754.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [TRAIN] Epoch: 4: 100%|██████████| 314/314 [03:25<00:00,  1.53it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS TRAIN: 0.709 || ACC TRAIN: 78.22\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [VAL] Epoch: 4: 100%|██████████| 50/50 [00:30<00:00,  1.62it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS VAL: 1.311 || ACC VAL: 62.20\n",
      "!!Saving model: /scratch/rw2268/Handbag_brand/snapshot/ResNet18/0005.pth!\n",
      "Decay learning rate to: 0.0009193985326646091.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [TRAIN] Epoch: 5: 100%|██████████| 314/314 [03:24<00:00,  1.53it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS TRAIN: 0.610 || ACC TRAIN: 81.07\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [VAL] Epoch: 5: 100%|██████████| 50/50 [00:30<00:00,  1.63it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS VAL: 1.574 || ACC VAL: 58.41\n",
      "Decay learning rate to: 0.0009040752237868656.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [TRAIN] Epoch: 6: 100%|██████████| 314/314 [03:24<00:00,  1.35it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS TRAIN: 0.476 || ACC TRAIN: 85.56\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [VAL] Epoch: 6: 100%|██████████| 50/50 [00:30<00:00,  1.62it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS VAL: 1.316 || ACC VAL: 65.11\n",
      "!!Saving model: /scratch/rw2268/Handbag_brand/snapshot/ResNet18/0007.pth!\n",
      "Decay learning rate to: 0.0008890073033904179.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [TRAIN] Epoch: 7: 100%|██████████| 314/314 [03:25<00:00,  1.53it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS TRAIN: 0.413 || ACC TRAIN: 87.55\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [VAL] Epoch: 7: 100%|██████████| 50/50 [00:30<00:00,  1.63it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS VAL: 1.638 || ACC VAL: 60.43\n",
      "Decay learning rate to: 0.0008741905150005775.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [TRAIN] Epoch: 8: 100%|██████████| 314/314 [03:25<00:00,  2.14it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS TRAIN: 0.333 || ACC TRAIN: 89.67\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [VAL] Epoch: 8: 100%|██████████| 50/50 [00:30<00:00,  2.33it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS VAL: 1.418 || ACC VAL: 63.84\n",
      "Decay learning rate to: 0.0008596206730839012.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [TRAIN] Epoch: 9: 100%|██████████| 314/314 [03:24<00:00,  2.58it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS TRAIN: 0.283 || ACC TRAIN: 90.69\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "!RESNET18 -> [VAL] Epoch: 9: 100%|██████████| 50/50 [00:30<00:00,  1.71it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LOSS VAL: 1.469 || ACC VAL: 65.11\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "train('ResNet18',train_loader, val_loader, model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "device(type='cuda', index=0)"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "device"
   ]
  },
  {
   "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": "3.5.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}