{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "3035f8a5-3aa6-4b11-8507-1e61e846fe23",
"metadata": {
"tags": []
},
"outputs": [],
"source": [
"!pip install -r requirements.txt"
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "e098af0e-21cb-42ce-84bd-fcb49e5d7f81",
"metadata": {
"tags": []
},
"outputs": [],
"source": [
"import pathlib\n",
"import pandas as pd\n",
"import numpy as np\n",
"import requests\n",
"import cv2\n",
"from tqdm.notebook import tqdm\n",
"import re\n",
"from PIL import Image"
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "7548cbea-2330-45d7-91b0-7f8644b0e6a0",
"metadata": {
"tags": []
},
"outputs": [],
"source": [
"def get_content(url):\n",
" retries = 0\n",
" while retries < 3:\n",
" try:\n",
" resp = requests.get(url)\n",
" if resp.status_code == 200:\n",
" return resp.content\n",
" else:\n",
" retries += 1\n",
" except requests.exceptions.Timeout:\n",
" retries += 1\n",
" except requests.exceptions.TooManyRedirects as e:\n",
" raise SystemExit(e)\n",
" except requests.exceptions.RequestException as e:\n",
" raise SystemExit(e)\n",
"\n",
"def get_category(entry):\n",
" if entry in ['A', 'B', 'C']:\n",
" return entry\n",
" else:\n",
" return 'N'\n",
"\n",
"def display_image_from_series(df, num_images=16):\n",
" images_so_far = 0\n",
" cols = 8\n",
" rows = int(np.ceil(num_images/cols))\n",
" fig = plt.figure(figsize=(int(cols*2.5),int(rows*2.5)))\n",
" \n",
" for i, (filename, year) in enumerate(df[['filename', 'years']].values):\n",
" images_so_far += 1\n",
" ax = plt.subplot(rows, cols, images_so_far)\n",
" ax.axis('off')\n",
" ax.set_title(i)\n",
" image = Image.open(f'img/ABO/{year}/no_category/{filename}')\n",
" plt.imshow(image)\n",
" if images_so_far == num_images:\n",
" return"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "bff4b449-2875-4563-a6d5-fe1c349c87f0",
"metadata": {},
"outputs": [],
"source": [
"pred_1501_1549 = pd.read_csv('data/predictions/pred_1501_1549.csv')\n",
"pred_1550_1599 = pd.read_csv('data/predictions/pred_1550_1599.csv')\n",
"pred_1600_1650 = pd.read_csv('data/predictions/pred_1600_1650.csv')\n",
"pred_1651_1699 = pd.read_csv('data/predictions/pred_1651_1699.csv')\n",
"pred_1700_1738 = pd.read_csv('data/predictions/pred_1700_1738.csv')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "923560af-a53f-4ae0-b37c-17793d298d4f",
"metadata": {},
"outputs": [],
"source": [
"pred_1501_1549['years'] = '1501_1549'\n",
"pred_1550_1599['years'] = '1550_1599'\n",
"pred_1600_1650['years'] = '1600_1650'\n",
"pred_1651_1699['years'] = '1651_1699'\n",
"pred_1700_1738['years'] = '1700_1738'\n",
"\n",
"all_frames = [pred_1501_1549, pred_1550_1599, pred_1600_1650, pred_1651_1699, pred_1700_1738]\n",
"all_pred = pd.concat(all_frames)\n",
"all_pred.to_csv('data/predictions/all_pred_1501_1738.csv')\n",
"print(len(all_pred))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "25bd6f96-6386-461a-9e6d-f86b12ab824b",
"metadata": {},
"outputs": [],
"source": [
"all_pred['probability'] = all_pred['probability'].apply(lambda x: json.loads(x))\n",
"all_pred['p_A'] = all_pred['probability'].apply(lambda x: x[0])\n",
"all_pred['p_B'] = all_pred['probability'].apply(lambda x: x[1])\n",
"all_pred['p_C'] = all_pred['probability'].apply(lambda x: x[2])\n",
"all_pred['p_N'] = all_pred['probability'].apply(lambda x: x[3])"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a96c3309-0f50-4d00-b321-e87245d49ef5",
"metadata": {},
"outputs": [],
"source": [
"def create_GT_addition(GT_df, candidate_df, num_additions):\n",
" return_lis = []\n",
" running_index = GT_df.iloc[-1]['Unnamed: 0']\n",
" for i, row in enumerate(candidate_df.values):\n",
" if len(return_lis) == num_additions:\n",
" print('Finishing at index', i)\n",
" break\n",
" bc = row[0].split('_')[0]\n",
" if GT_df['Strichcode'].str.contains(bc).any():\n",
" continue\n",
" page = row[0].split('_')[1].replace('.jpg', '')\n",
" running_index += 1\n",
" new_row = {\n",
" 'Unnamed: 0': running_index,\n",
" 'Strichcode': bc,\n",
" 'Link': '',\n",
" 'Variante': row[1],\n",
" 'Farbe': '',\n",
" 'Erhaltungsgrad': '',\n",
" 'Bem.': '',\n",
" 'Seite': int(page),\n",
" 'Image URLs': [f'https://iiif.onb.ac.at/images/ABO/{bc}/{page}/full/full/0/native.jpg']\n",
" }\n",
" return_lis.append(new_row)\n",
" return pd.DataFrame(return_lis)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "830fe823-75c0-4e95-9dc5-55a7fedede28",
"metadata": {},
"outputs": [],
"source": [
"GT_df = pd.read_csv('data/groundtruth/BE_GT_v2.csv')\n",
"\n",
"GT_addition_A = create_GT_addition(GT_df, all_pred[all_pred['p_A'] > 0.95], 79)\n",
"GT_addition_C = create_GT_addition(GT_df, all_pred[all_pred['p_C'] > 0.95], 150)\n",
"new_GT = pd.concat([GT_df, GT_addition_A, GT_addition_C], ignore_index=True)\n",
"new_GT = new_GT.drop('Unnamed: 0', axis=1)\n",
"new_GT.to_csv('data/groundtruth/BE_GT_v3.csv')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "94ccf8d2-dd0a-469e-9f94-349ab546deda",
"metadata": {},
"outputs": [],
"source": [
"display_image_from_series(all_pred[all_pred['p_A'] > 0.95], num_images=85)"
]
},
{
"cell_type": "markdown",
"id": "18c4f053-3b42-4ee7-97d4-91e8e71570b4",
"metadata": {},
"source": [
"## Download new GT with square cutout and reduced resolution"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "3cad43ef-7c46-4949-ae60-5afd06a7bb80",
"metadata": {
"tags": []
},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "acf5edc590724e54a41c72adbcc97ad5",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
" 0%| | 0/804 [00:00<?, ?it/s]"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"img_dir = pathlib.Path('img')\n",
"if not img_dir.exists():\n",
" img_dir.mkdir()\n",
"GT_path = img_dir.joinpath('GT_square')\n",
"if not GT_path.exists():\n",
" GT_path.mkdir()\n",
"GT_csv = pd.read_csv('data/groundtruth/BE_GT_v3.csv')\n",
"for category, URL_lis, barcode in zip(tqdm(GT_csv['Variante']), GT_csv['Image URLs'], GT_csv['Strichcode']):\n",
" url_lis = eval(URL_lis)\n",
" cat = get_category(category)\n",
" GT_cat_path = GT_path.joinpath(cat)\n",
" if not GT_cat_path.exists():\n",
" GT_cat_path.mkdir()\n",
" for url in url_lis:\n",
" page_number = re.findall('/(.{8,12})/full/full', url)[0]\n",
" filename = f'{barcode}_{page_number}.jpg'\n",
" if 'REPO' not in url:\n",
" filepath = GT_cat_path.joinpath(filename)\n",
" if not filepath.exists():\n",
" url = url.replace('full/full', 'square/256,')\n",
" img_content = get_content(url)\n",
" open(filepath, 'wb').write(img_content)\n",
" else:\n",
" filepath = GT_cat_path.joinpath(filename.replace('.jpg.', '.'))\n",
" if not filepath.exists():\n",
" resp = requests.get(url, stream=True).raw\n",
" img = np.asarray(bytearray(resp.read()), dtype='uint8')\n",
" img = cv2.imdecode(img, cv2.IMREAD_COLOR)\n",
" img_width = img.shape[1]\n",
" img_height = img.shape[0]\n",
" x = 0\n",
" y = int((img_height - img_width)/2)\n",
" w = img_width\n",
" h = img_width\n",
" square_img_content = img[y:y+h, x:x+w]\n",
" resized_img = cv2.resize(square_img_content, (256, 256))\n",
" cv2.imwrite(filepath.as_posix(), resized_img)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "dd1ab4b1-e938-4337-856b-f04ae7b367ac",
"metadata": {},
"outputs": [],
"source": []
}
],
"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": "3.10.10"
}
},
"nbformat": 4,
"nbformat_minor": 5
}