final commit

Author: nsk126

README.md

@@ -1 +1,7 @@
-# transferlearning_shufflenet
+# Transfer Learning on ShuffleNetv2
+
+- A few CNN comparisions using Tensorflow on the CIFAR-10 dataset. Show progessive improvement with use of maxpool & dropout.
+[CNN comparisions](./main.ipynb)
+
+- Transferlearning on the ShuffleNetv2 model in Pytorch.
+[Transferlearning on ShuffleNet](./shuffle.ipynb)

main.ipynb

@@ -1,5 +1,17 @@
 {
  "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# CNNs on CIFAR-10\n",
+    "\n",
+    "Comparing different CNNs performance on the CIFAR-10 Dataset\n",
+    "- Just CNNs\n",
+    "- Include MaxPool\n",
+    "- Include Dropout"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 2,

shuffle.ipynb

@@ -1,5 +1,14 @@
 {
  "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Tranfer Learning over Shufflnetv2\n",
+    "\n",
+    "Fine Tuning a Shufflenet model (pre-trained over Imagenet) on the CIFAR-10 Dataset."
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 2,
@@ -18,6 +27,13 @@
     "from torchsummary import summary\n"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "#### Training, Valiation Loop"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 13,
@@ -47,7 +63,7 @@
     "                outputs = model(inputs) # prediction\n",
     "                _loss = loss(outputs, labels) # loss\n",
     "                _, pred = torch.max(outputs, 1)\n",
-    "                \n",
+    "\n",
     "                # backpropr\n",
     "                _loss.backward()\n",
     "                optim.step()\n",
@@ -83,11 +99,11 @@
     "            print(f'Epoch #{epoch+1}, '\n",
     "                    f'Validation Loss: {valid_loss/len(testload.dataset):.2f}, '\n",
     "                    f'Validation Acc: {valid_acc.double()/len(testload.dataset):.2f}')\n",
-    "            \n",
+    "\n",
     "        # Save losses for plotting\n",
     "        train_losses.append(running_loss/len(trainload.dataset))\n",
     "        valid_losses.append(valid_loss/len(testload.dataset))\n",
-    "            \n",
+    "\n",
     "    print(f'Training time: {start_time - time()}')\n",
     "    print('Savin model')\n",
     "    torch.save(model.state_dict(), 'my_shufflenet.pt')\n",
@@ -314,6 +330,15 @@
     "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "#### Freeze network params\n",
+    "- Freeze all params in the network\n",
+    "- Replace the final classification layer with a new one to train."
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 10,
@@ -330,6 +355,13 @@
     "model = model.to(device)"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "#### Default transformations"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 11,
@@ -357,6 +389,13 @@
     "])"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "#### Dataloaders"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 12,