Skip to content

Commit

Permalink
Added basic CNNs
Browse files Browse the repository at this point in the history
  • Loading branch information
@alvanli
alvanli committed May 19, 2022
1 parent 5e0105b commit bbff537
Show file tree
Hide file tree
Showing 2 changed files with 553 additions and 0 deletions.
394 changes: 394 additions & 0 deletions CV/BaseCNN.ipynb
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,394 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "cddfec5d",
"metadata": {},
"outputs": [],
"source": [
"import matplotlib.pyplot as plt\n",
"import numpy as np\n",
"import os\n",
"import tensorflow as tf\n",
"from tensorflow.keras.utils import image_dataset_from_directory"
]
},
{
"cell_type": "code",
"execution_count": 27,
"id": "2bb91ad6",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Found 694 files belonging to 2 classes.\n",
"Found 200 files belonging to 2 classes.\n"
]
}
],
"source": [
"BATCH_SIZE = 32\n",
"NUM_CLASSES = 2\n",
"IMG_SIZE = 224\n",
"NUM_EPOCHS = 15\n",
"\n",
"TRAIN_DIR = \"C://Users//alvin//DATA//dsc//cxc-2022//data//alien_predator//train\"\n",
"TEST_DIR = \"C://Users//alvin//DATA//dsc//cxc-2022//data//alien_predator//validation\"\n",
"\n",
"train_ds = image_dataset_from_directory(TRAIN_DIR,\n",
" shuffle=True,\n",
" batch_size=BATCH_SIZE,\n",
" image_size=(IMG_SIZE, IMG_SIZE))\n",
"\n",
"test_ds = image_dataset_from_directory(TEST_DIR,\n",
" shuffle=True,\n",
" batch_size=BATCH_SIZE,\n",
" image_size=(IMG_SIZE, IMG_SIZE))\n",
" "
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "a6ac7ebc",
"metadata": {},
"outputs": [],
"source": [
"model = tf.keras.Sequential([\n",
" tf.keras.layers.Resizing(IMG_SIZE, IMG_SIZE),\n",
" tf.keras.layers.Rescaling(1./255),\n",
" tf.keras.layers.Conv2D(16, 3, padding='same', activation='relu'),\n",
" tf.keras.layers.MaxPooling2D(),\n",
" tf.keras.layers.Conv2D(32, 3, padding='same', activation='relu'),\n",
" tf.keras.layers.MaxPooling2D(),\n",
" tf.keras.layers.Conv2D(64, 3, padding='same', activation='relu'),\n",
" tf.keras.layers.MaxPooling2D(),\n",
" tf.keras.layers.Flatten(),\n",
" tf.keras.layers.Dense(128, activation='relu'),\n",
" tf.keras.layers.Dense(NUM_CLASSES)\n",
"])"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "8d01f717",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Epoch 1/15\n",
"22/22 [==============================] - 9s 405ms/step - loss: 1.2455 - acc: 0.5274 - val_loss: 0.6716 - val_acc: 0.6100\n",
"Epoch 2/15\n",
"22/22 [==============================] - 10s 435ms/step - loss: 0.6139 - acc: 0.6470 - val_loss: 0.6061 - val_acc: 0.6450\n",
"Epoch 3/15\n",
"22/22 [==============================] - 10s 450ms/step - loss: 0.5307 - acc: 0.7522 - val_loss: 0.6285 - val_acc: 0.6500\n",
"Epoch 4/15\n",
"22/22 [==============================] - 10s 444ms/step - loss: 0.4701 - acc: 0.7997 - val_loss: 0.5551 - val_acc: 0.7200\n",
"Epoch 5/15\n",
"22/22 [==============================] - 10s 438ms/step - loss: 0.3152 - acc: 0.8833 - val_loss: 0.5526 - val_acc: 0.7300\n",
"Epoch 6/15\n",
"22/22 [==============================] - 12s 536ms/step - loss: 0.2028 - acc: 0.9265 - val_loss: 0.7496 - val_acc: 0.7050\n",
"Epoch 7/15\n",
"22/22 [==============================] - 17s 753ms/step - loss: 0.1850 - acc: 0.9251 - val_loss: 0.7892 - val_acc: 0.6500\n",
"Epoch 8/15\n",
"22/22 [==============================] - 15s 677ms/step - loss: 0.1292 - acc: 0.9640 - val_loss: 1.0998 - val_acc: 0.6900\n",
"Epoch 9/15\n",
"22/22 [==============================] - 16s 729ms/step - loss: 0.0668 - acc: 0.9798 - val_loss: 0.9719 - val_acc: 0.7000\n",
"Epoch 10/15\n",
"22/22 [==============================] - 16s 734ms/step - loss: 0.0348 - acc: 0.9942 - val_loss: 1.0952 - val_acc: 0.6950\n",
"Epoch 11/15\n",
"22/22 [==============================] - 16s 718ms/step - loss: 0.0164 - acc: 0.9986 - val_loss: 1.3058 - val_acc: 0.7200\n",
"Epoch 12/15\n",
"22/22 [==============================] - 16s 718ms/step - loss: 0.0145 - acc: 0.9971 - val_loss: 1.3329 - val_acc: 0.7200\n",
"Epoch 13/15\n",
"22/22 [==============================] - 16s 725ms/step - loss: 0.0111 - acc: 0.9986 - val_loss: 1.4546 - val_acc: 0.7000\n",
"Epoch 14/15\n",
"22/22 [==============================] - 16s 732ms/step - loss: 0.0036 - acc: 1.0000 - val_loss: 1.6123 - val_acc: 0.7050\n",
"Epoch 15/15\n",
"22/22 [==============================] - 16s 735ms/step - loss: 0.0026 - acc: 1.0000 - val_loss: 1.6705 - val_acc: 0.7000\n"
]
}
],
"source": [
"model.compile(\n",
" optimizer=tf.keras.optimizers.Adam(),\n",
" loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
" metrics=['acc']\n",
")\n",
"\n",
"\n",
"history = model.fit(\n",
" train_ds,\n",
" validation_data=test_ds,\n",
" epochs=NUM_EPOCHS\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "7208139f",
"metadata": {},
"outputs": [],
"source": [
"# talk about overfitting, now adding dropout"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "041bd4a4",
"metadata": {},
"outputs": [],
"source": [
"dropout_model = tf.keras.Sequential([\n",
" tf.keras.layers.Resizing(IMG_SIZE, IMG_SIZE),\n",
" tf.keras.layers.Rescaling(1./255),\n",
" tf.keras.layers.Conv2D(16, 3, padding='same', activation='relu'),\n",
" tf.keras.layers.MaxPooling2D(),\n",
" tf.keras.layers.Dropout(0.3),\n",
" tf.keras.layers.Conv2D(32, 3, padding='same', activation='relu'),\n",
" tf.keras.layers.MaxPooling2D(),\n",
" tf.keras.layers.Conv2D(64, 3, padding='same', activation='relu'),\n",
" tf.keras.layers.MaxPooling2D(),\n",
" tf.keras.layers.Dropout(0.3),\n",
" tf.keras.layers.Flatten(),\n",
" tf.keras.layers.Dense(128, activation='relu'),\n",
" tf.keras.layers.Dense(NUM_CLASSES)\n",
"])"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "dff898ba",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Epoch 1/15\n",
"22/22 [==============================] - 19s 841ms/step - loss: 1.0004 - acc: 0.4755 - val_loss: 0.6842 - val_acc: 0.5400\n",
"Epoch 2/15\n",
"22/22 [==============================] - 19s 861ms/step - loss: 0.6597 - acc: 0.5461 - val_loss: 0.6651 - val_acc: 0.6650\n",
"Epoch 3/15\n",
"22/22 [==============================] - 18s 822ms/step - loss: 0.5992 - acc: 0.6830 - val_loss: 0.6338 - val_acc: 0.6550\n",
"Epoch 4/15\n",
"22/22 [==============================] - 19s 866ms/step - loss: 0.5560 - acc: 0.7478 - val_loss: 0.5589 - val_acc: 0.6850\n",
"Epoch 5/15\n",
"22/22 [==============================] - 19s 846ms/step - loss: 0.4456 - acc: 0.7968 - val_loss: 0.5684 - val_acc: 0.6950\n",
"Epoch 6/15\n",
"22/22 [==============================] - 19s 869ms/step - loss: 0.4187 - acc: 0.8516 - val_loss: 0.5490 - val_acc: 0.7200\n",
"Epoch 7/15\n",
"22/22 [==============================] - 19s 860ms/step - loss: 0.2824 - acc: 0.8818 - val_loss: 0.5506 - val_acc: 0.7100\n",
"Epoch 8/15\n",
"22/22 [==============================] - 20s 894ms/step - loss: 0.2260 - acc: 0.9222 - val_loss: 0.5818 - val_acc: 0.7000\n",
"Epoch 9/15\n",
"22/22 [==============================] - 20s 895ms/step - loss: 0.1209 - acc: 0.9640 - val_loss: 0.6214 - val_acc: 0.7200\n",
"Epoch 10/15\n",
"22/22 [==============================] - 19s 870ms/step - loss: 0.0976 - acc: 0.9741 - val_loss: 0.7811 - val_acc: 0.6700\n",
"Epoch 11/15\n",
"22/22 [==============================] - 20s 907ms/step - loss: 0.0605 - acc: 0.9813 - val_loss: 0.8814 - val_acc: 0.7250\n",
"Epoch 12/15\n",
"22/22 [==============================] - 19s 862ms/step - loss: 0.0546 - acc: 0.9798 - val_loss: 0.9069 - val_acc: 0.7300\n",
"Epoch 13/15\n",
"22/22 [==============================] - 20s 890ms/step - loss: 0.0386 - acc: 0.9914 - val_loss: 0.9644 - val_acc: 0.6900\n",
"Epoch 14/15\n",
"22/22 [==============================] - 20s 884ms/step - loss: 0.0475 - acc: 0.9827 - val_loss: 0.9680 - val_acc: 0.6850\n",
"Epoch 15/15\n",
"22/22 [==============================] - 20s 905ms/step - loss: 0.0415 - acc: 0.9870 - val_loss: 1.0254 - val_acc: 0.7500\n"
]
}
],
"source": [
"dropout_model.compile(\n",
" optimizer=tf.keras.optimizers.Adam(),\n",
" loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
" metrics=['acc']\n",
")\n",
"\n",
"\n",
"dropout_history = dropout_model.fit(\n",
" train_ds,\n",
" validation_data=test_ds,\n",
" epochs=NUM_EPOCHS\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "01064e0f",
"metadata": {},
"outputs": [],
"source": [
"# still overfitting haha"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "11962da7",
"metadata": {},
"outputs": [],
"source": [
"# Might be good to introduce ReduceLRonPlateau and EarlyStopping here too"
]
},
{
"cell_type": "code",
"execution_count": 29,
"id": "0c98b8b5",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Found 694 images belonging to 2 classes.\n",
"Found 200 images belonging to 2 classes.\n"
]
}
],
"source": [
"train_aug_ds = tf.keras.preprocessing.image.ImageDataGenerator(\n",
" brightness_range=[0.9, 1.1],\n",
" rotation_range=30,\n",
" width_shift_range=0.1,\n",
" height_shift_range=0.1,\n",
" shear_range=0.1,\n",
" zoom_range=0.1,\n",
" fill_mode=\"nearest\",\n",
" horizontal_flip=True,\n",
" vertical_flip=True,\n",
" rescale=1.0/255.0\n",
").flow_from_directory(TRAIN_DIR, batch_size=BATCH_SIZE, target_size=(IMG_SIZE, IMG_SIZE), shuffle=True, class_mode='sparse')\n",
"\n",
"\n",
"test_aug_ds = tf.keras.preprocessing.image.ImageDataGenerator(rescale=1.0/255.0).flow_from_directory(\n",
" TEST_DIR, batch_size=BATCH_SIZE, target_size=(IMG_SIZE, IMG_SIZE), shuffle=True, class_mode='sparse'\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 32,
"id": "5c30af59",
"metadata": {},
"outputs": [],
"source": [
"del dropout_model\n",
"\n",
"dropout_model = tf.keras.Sequential([\n",
" tf.keras.layers.Conv2D(16, 3, padding='same', activation='relu'),\n",
" tf.keras.layers.MaxPooling2D(),\n",
" tf.keras.layers.Dropout(0.3),\n",
" tf.keras.layers.Conv2D(32, 3, padding='same', activation='relu'),\n",
" tf.keras.layers.MaxPooling2D(),\n",
" tf.keras.layers.Conv2D(64, 3, padding='same', activation='relu'),\n",
" tf.keras.layers.MaxPooling2D(),\n",
" tf.keras.layers.Dropout(0.3),\n",
" tf.keras.layers.Flatten(),\n",
" tf.keras.layers.Dense(128, activation='relu'),\n",
" tf.keras.layers.Dense(NUM_CLASSES)\n",
"])"
]
},
{
"cell_type": "code",
"execution_count": 33,
"id": "c892e485",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Epoch 1/20\n",
"22/22 [==============================] - 22s 978ms/step - loss: 1.4652 - acc: 0.4597 - val_loss: 0.6885 - val_acc: 0.5050\n",
"Epoch 2/20\n",
"22/22 [==============================] - 22s 988ms/step - loss: 0.6711 - acc: 0.5778 - val_loss: 0.6485 - val_acc: 0.6100\n",
"Epoch 3/20\n",
"22/22 [==============================] - 22s 988ms/step - loss: 0.6291 - acc: 0.6455 - val_loss: 0.6165 - val_acc: 0.6450\n",
"Epoch 4/20\n",
"22/22 [==============================] - 22s 990ms/step - loss: 0.5852 - acc: 0.6974 - val_loss: 0.7310 - val_acc: 0.5700\n",
"Epoch 5/20\n",
"22/22 [==============================] - 22s 1s/step - loss: 0.6158 - acc: 0.6657 - val_loss: 0.6325 - val_acc: 0.6400\n",
"Epoch 6/20\n",
"22/22 [==============================] - 22s 977ms/step - loss: 0.6196 - acc: 0.6671 - val_loss: 0.6182 - val_acc: 0.6500\n",
"Epoch 7/20\n",
"22/22 [==============================] - 22s 975ms/step - loss: 0.5712 - acc: 0.7075 - val_loss: 0.6478 - val_acc: 0.5900\n",
"Epoch 8/20\n",
"22/22 [==============================] - 22s 971ms/step - loss: 0.5752 - acc: 0.7032 - val_loss: 0.7221 - val_acc: 0.5600\n",
"Epoch 9/20\n",
"22/22 [==============================] - 21s 958ms/step - loss: 0.5577 - acc: 0.7248 - val_loss: 0.5665 - val_acc: 0.6950\n",
"Epoch 10/20\n",
"22/22 [==============================] - 23s 1s/step - loss: 0.5383 - acc: 0.7248 - val_loss: 0.5474 - val_acc: 0.7050\n",
"Epoch 11/20\n",
"22/22 [==============================] - 24s 1s/step - loss: 0.5226 - acc: 0.7450 - val_loss: 0.6188 - val_acc: 0.6400\n",
"Epoch 12/20\n",
"22/22 [==============================] - 22s 988ms/step - loss: 0.5254 - acc: 0.7450 - val_loss: 0.5570 - val_acc: 0.7300\n",
"Epoch 13/20\n",
"22/22 [==============================] - 22s 989ms/step - loss: 0.5132 - acc: 0.7522 - val_loss: 0.5408 - val_acc: 0.7200\n",
"Epoch 14/20\n",
"22/22 [==============================] - 22s 969ms/step - loss: 0.5049 - acc: 0.7536 - val_loss: 0.5888 - val_acc: 0.6800\n",
"Epoch 15/20\n",
"22/22 [==============================] - 21s 966ms/step - loss: 0.5174 - acc: 0.7392 - val_loss: 0.5437 - val_acc: 0.7300\n",
"Epoch 16/20\n",
"22/22 [==============================] - 22s 960ms/step - loss: 0.4946 - acc: 0.7695 - val_loss: 0.5308 - val_acc: 0.7150\n",
"Epoch 17/20\n",
"22/22 [==============================] - 22s 990ms/step - loss: 0.4957 - acc: 0.7622 - val_loss: 0.5240 - val_acc: 0.7450\n",
"Epoch 18/20\n",
"22/22 [==============================] - 14s 598ms/step - loss: 0.5104 - acc: 0.7565 - val_loss: 0.6348 - val_acc: 0.6900\n",
"Epoch 19/20\n",
"22/22 [==============================] - 12s 531ms/step - loss: 0.4926 - acc: 0.7695 - val_loss: 0.6126 - val_acc: 0.6900\n",
"Epoch 20/20\n",
"22/22 [==============================] - 12s 527ms/step - loss: 0.4826 - acc: 0.7911 - val_loss: 0.5567 - val_acc: 0.7300\n"
]
}
],
"source": [
"dropout_model.compile(\n",
" optimizer=tf.keras.optimizers.Adam(),\n",
" loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
" metrics=['acc']\n",
")\n",
"\n",
"\n",
"dropout_history = dropout_model.fit(\n",
" train_aug_ds,\n",
" validation_data=test_aug_ds,\n",
" epochs=NUM_EPOCHS + 5\n",
")"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"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": "3.8.8"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
Loading

0 comments on commit bbff537

Please sign in to comment.