diff --git a/classification_model_training/files_from_makeathon/data/test.dvc b/classification_model_training/files_from_makeathon/data/test.dvc index f6f370a..5cc43a9 100644 --- a/classification_model_training/files_from_makeathon/data/test.dvc +++ b/classification_model_training/files_from_makeathon/data/test.dvc @@ -1,6 +1,6 @@ outs: -- md5: e03ff535d9f3da7f51ed642b70ec5d95.dir - size: 549163 +- md5: c6f373f6997ea2bae313d494caa35b43.dir + size: 1434598 nfiles: 80 hash: md5 path: test diff --git a/classification_model_training/files_from_makeathon/data/train.dvc b/classification_model_training/files_from_makeathon/data/train.dvc index 5df3ee2..25b2feb 100644 --- a/classification_model_training/files_from_makeathon/data/train.dvc +++ b/classification_model_training/files_from_makeathon/data/train.dvc @@ -1,6 +1,6 @@ outs: -- md5: 0263429cb9fefbc8b68b6a88c26d942f.dir - size: 2506709 +- md5: c6c8f93cb47471d0133562e34c3f56d0.dir + size: 6053320 nfiles: 362 hash: md5 path: train diff --git a/classification_model_training/files_from_makeathon/dvc.lock b/classification_model_training/files_from_makeathon/dvc.lock index 70c6532..c34fb9b 100644 --- a/classification_model_training/files_from_makeathon/dvc.lock +++ b/classification_model_training/files_from_makeathon/dvc.lock @@ -97,3 +97,47 @@ stages: hash: md5 md5: 49a445cb95476dbc8dc7e7e2530050f0 size: 74 + evaluate-last: + cmd: python predict.py --model efficientnet-b2 --image_size 224 --num_classes + 4 --dataset ./data/test/images --gt ./data/test/gt.csv --single_model_path models/model.pt + deps: + - path: data/test + hash: md5 + md5: e03ff535d9f3da7f51ed642b70ec5d95.dir + size: 549163 + nfiles: 80 + - path: models/model.pt + hash: md5 + md5: b7e4e7f26989ba76d1009d1a78724b9c + size: 31270857 + - path: predict.py + hash: md5 + md5: 43fb163a6a3546d1bd6b47cf92fc6ff5 + size: 6001 + params: + params.yaml: + base: + random_seed: 1 + evaluate: + dataset: ./data/test/images + gt: ./data/test/gt.csv + device: cuda + val_bs: 16 + train: + fold: 1 + dataset: ./data/train/images + gt: ./data/train/gt.csv + num_classes: 4 + metric: f1_score + device: cuda + epochs: 30 + train_batch: 16 + val_batch: 16 + outdir: models + lr: 0.001 + dropout_rate: 0.3 + drop_connect_rate: 0.2 + batch_norm_momentum: 0.99 + batch_norm_epsilon: 0.001 + model: efficientnet-b2 + image_size: 224 diff --git a/classification_model_training/files_from_makeathon/dvc.yaml b/classification_model_training/files_from_makeathon/dvc.yaml index 97ba0b2..6560fb7 100644 --- a/classification_model_training/files_from_makeathon/dvc.yaml +++ b/classification_model_training/files_from_makeathon/dvc.yaml @@ -59,3 +59,20 @@ stages: metrics: - models/metrics-evaluate.json: cache: false + evaluate-last: + cmd: >- + python predict.py + --model ${train.model} + --image_size ${train.image_size} + --num_classes ${train.num_classes} + --dataset ${evaluate.dataset} + --gt ${evaluate.gt} + --single_model_path ${train.outdir}/model.pt + deps: + - data/test + - predict.py + - models/model.pt + params: + - base + - train + - evaluate diff --git a/classification_model_training/files_from_makeathon/notebooks/crop.ipynb b/classification_model_training/files_from_makeathon/notebooks/crop.ipynb new file mode 100644 index 0000000..bd2cc7b --- /dev/null +++ b/classification_model_training/files_from_makeathon/notebooks/crop.ipynb @@ -0,0 +1,347 @@ +{ + "cells": [ + { + "cell_type": "code", + "execution_count": 1, + "id": "cc5ebda6-6120-47c3-8240-b74fe83b16b4", + "metadata": {}, + "outputs": [], + "source": [ + "from pathlib import Path\n", + "import numpy as np\n", + "import pandas as pd\n", + "import seaborn as sns\n", + "\n", + "import cv2\n", + "import matplotlib.pyplot as plt" + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "id": "7c6bf02f-8330-4509-b489-0f0f0eca787b", + "metadata": {}, + "outputs": [], + "source": [ + "csv_train_path = '../data/train/gt.csv'\n", + "csv_test_path = '../data/test/gt.csv'" + ] + }, + { + "cell_type": "code", + "execution_count": 3, + "id": "124bb387-ed02-461e-8e8e-080229fecc86", + "metadata": {}, + "outputs": [], + "source": [ + "df_train = pd.read_csv(csv_train_path)\n", + "df_train['split'] = 'train'\n", + "df_test = pd.read_csv(csv_test_path)\n", + "df_test['split'] = 'test'" + ] + }, + { + "cell_type": "code", + "execution_count": 4, + "id": "cbe2f50e-2407-4aa4-b957-e3426681fde9", + "metadata": {}, + "outputs": [], + "source": [ + "label_map = {\n", + " 0:\"positive\",\n", + " 1:\"negative\",\n", + " 2:\"empty\",\n", + " 3:\"invalid\",\n", + "}\n", + "\n", + "split_map = {\n", + " 0:\"train\",\n", + " 1:\"val\",\n", + " -1:\"test\",\n", + "}" + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "id": "5739a88f-739b-4ffd-9ee1-843aac5b5559", + "metadata": {}, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
imagetargetfoldsplitbrandstem
0safecare_negative_030_jpg.rf.7c617838cb5f06ef9...10trainsafecaresafecare_negative_030
1safecare_invalid_swap_038_jpg.rf.89441e6b07934...30trainsafecaresafecare_invalid_swap_038
2hygisun_negative_005_jpg.rf.ff1b59ca7eb97855c3...10trainhygisunhygisun_negative_005
3medicovid_negative_real_005_png.rf.8502b610e6f...10trainmedicovidmedicovid_negative_real_005
4safecare_invalid_swap_012_jpg.rf.9a46f6916090f...30trainsafecaresafecare_invalid_swap_012
\n", + "
" + ], + "text/plain": [ + " image target fold split \n", + "0 safecare_negative_030_jpg.rf.7c617838cb5f06ef9... 1 0 train \\\n", + "1 safecare_invalid_swap_038_jpg.rf.89441e6b07934... 3 0 train \n", + "2 hygisun_negative_005_jpg.rf.ff1b59ca7eb97855c3... 1 0 train \n", + "3 medicovid_negative_real_005_png.rf.8502b610e6f... 1 0 train \n", + "4 safecare_invalid_swap_012_jpg.rf.9a46f6916090f... 3 0 train \n", + "\n", + " brand stem \n", + "0 safecare safecare_negative_030 \n", + "1 safecare safecare_invalid_swap_038 \n", + "2 hygisun hygisun_negative_005 \n", + "3 medicovid medicovid_negative_real_005 \n", + "4 safecare safecare_invalid_swap_012 " + ] + }, + "execution_count": 5, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "df_cls = pd.concat([df_train,df_test])\n", + "df_cls['brand'] = df_cls['image'].str.extract(r'([^_]+)')\n", + "df_cls['stem'] = df_cls.image.str.extract('(.*)_[jpg]')\n", + "df_cls['fold'] = df_cls['fold'].fillna(-1).astype('int')\n", + "df_cls.head()" + ] + }, + { + "cell_type": "code", + "execution_count": 6, + "id": "f7061f91-41f3-4de8-8aba-1dbd2a3a3036", + "metadata": {}, + "outputs": [], + "source": [ + "raw_path = f\"../../../../covisionapp_image_upload/\"\n", + "stems = list((p.resolve(),p.resolve().stem) for p in Path(raw_path).glob(\"**/*\") if p.suffix in {\".jpg\", \".png\"})\n", + "df_raw = pd.DataFrame(stems)\n", + "df_raw.columns = ['raw_image', 'stem']\n", + "# df_raw.head()" + ] + }, + { + "cell_type": "code", + "execution_count": 7, + "id": "0510bfb9-6832-4b3f-9d48-33869be9263f", + "metadata": {}, + "outputs": [], + "source": [ + "txt_path = '../../../covision-training/data/raw/'\n", + "txts = list((p.resolve(),p.resolve().stem) for p in Path(txt_path).glob(\"**/labels/*\") if p.suffix in {\".txt\",})\n", + "df_txt = pd.DataFrame(txts)\n", + "df_txt.columns = ['raw_txt', 'stem']\n", + "df_txt['stem'] = df_txt['stem'].str.extract('(.*)_[jpg]')\n", + "# df_txt.head()" + ] + }, + { + "cell_type": "code", + "execution_count": 8, + "id": "e152d78a-438f-4854-927d-d94446d8f972", + "metadata": {}, + "outputs": [], + "source": [ + "# list(filter(lambda x: x=='safecare_negative_030', stems))" + ] + }, + { + "cell_type": "code", + "execution_count": 9, + "id": "bc70955f-cf23-42f9-87cf-f101c622588f", + "metadata": {}, + "outputs": [], + "source": [ + "assert len(df_cls[~df_cls['stem'].isin(df_raw['stem'])]) == 0" + ] + }, + { + "cell_type": "code", + "execution_count": 10, + "id": "4d2a0b0b-c865-45d4-95d2-1e04f58f8461", + "metadata": {}, + "outputs": [], + "source": [ + "assert len(df_cls[~df_cls['stem'].isin(df_txt['stem'])]) == 0" + ] + }, + { + "cell_type": "code", + "execution_count": 11, + "id": "5ad5c687-0bd2-48bc-ab9f-4691f968fc9d", + "metadata": {}, + "outputs": [], + "source": [ + "df = df_cls.merge(df_raw, on='stem').merge(df_txt, on='stem')\n", + "# df" + ] + }, + { + "cell_type": "code", + "execution_count": 12, + "id": "b94e822a-f9dd-46d3-95a9-b38e308a67cf", + "metadata": {}, + "outputs": [], + "source": [ + "root = Path('../data/')" + ] + }, + { + "cell_type": "code", + "execution_count": 13, + "id": "b33d839d-c448-4506-80ed-f564db3a6fe1", + "metadata": {}, + "outputs": [], + "source": [ + "df['new_image'] = df.raw_txt.map(lambda x: Path(x).stem + '.jpg')" + ] + }, + { + "cell_type": "code", + "execution_count": 14, + "id": "efae24b5-2811-482d-a081-71764306c2e7", + "metadata": {}, + "outputs": [], + "source": [ + "df[df['split']=='train'][['new_image','target','fold']].to_csv(root / 'train' / 'gt.csv', header=['image','target','fold'], index=False)" + ] + }, + { + "cell_type": "code", + "execution_count": 15, + "id": "4e8b239a-5a14-4ef7-ad03-ba3619bfce94", + "metadata": {}, + "outputs": [], + "source": [ + "df[df['split']=='test'][['new_image','target']].to_csv(root / 'test' / 'gt.csv', header=['image','target'], index=False)" + ] + }, + { + "cell_type": "code", + "execution_count": 16, + "id": "077831bb-bb4d-4250-ae97-7d5b390ece67", + "metadata": {}, + "outputs": [], + "source": [ + "maxwidth, maxheight = 300, 300\n", + "if True:\n", + " for row in df.itertuples():\n", + " img = cv2.imread(str(row.raw_image))\n", + " # plt.imshow(img)\n", + " with open(row.raw_txt) as f:\n", + " lb = np.array([x.split() for x in f.read().strip().splitlines()], dtype=np.float32)\n", + " \n", + " dh, dw = img.shape[:2]\n", + " x, y, w, h = lb[0][1:]\n", + " top = int((y - h / 2.) * dh)\n", + " left = int((x - w / 2.) * dw)\n", + " height = int(dh*h)\n", + " width = int(dw*w)\n", + " \n", + " cropped_image = img[top:top+height, left:left+width]\n", + " f1 = maxwidth / cropped_image.shape[1]\n", + " f2 = maxheight / cropped_image.shape[0]\n", + " f = min(f1, f2) # resizing factor\n", + " dim = (int(cropped_image.shape[1] * f), int(cropped_image.shape[0] * f))\n", + " cropped_image = cv2.resize(cropped_image, dim)\n", + " # plt.imshow(cropped_image)\n", + " \n", + " ds = 'train' if row.split in ['train','val'] else 'test'\n", + " fname = root / ds / 'images' / (row.raw_txt.stem + '.jpg')\n", + " # print(fname)\n", + " fname.parent.mkdir(parents=True, exist_ok=True)\n", + " cv2.imwrite(str(fname), cropped_image)" + ] + } + ], + "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": "3.9.17" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +}