{ "id": 223695, "title": "how to hand label images", "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/223695", "author_name": "hengck23", "post_date": "2021-03-05T06:17:20.342000", "votes": 13, "comment_count": 4, "views": 0, "content": "

i just label difficult cvc cases (endpoint only)

\n

\"\"
\n\"\"
\n\"\"

", "messages": [ { "id": 1227050, "postDate": "2021-03-05T06:17:20.343Z", "content": "

i just label difficult cvc cases (endpoint only)

\n

\"\"
\n\"\"
\n\"\"

", "rawMarkdown": "i just label difficult cvc cases (endpoint only)\n\n![](https://i.ibb.co/jb5vy1X/Selection-047.png)\n![](https://i.ibb.co/9YX7Y0d/Selection-046.png)\n![](https://i.ibb.co/VY6QJ6n/Selection-048.png)", "votes": 13 }, { "id": 1227209, "postDate": "2021-03-05T09:55:35.497Z", "content": "

Alternative approach is to use RFCx method:
\n-use model to generate high confidence candidates
\n-then use hand label to mark as FP and TP

", "rawMarkdown": "Alternative approach is to use RFCx method:\n-use model to generate high confidence candidates\n-then use hand label to mark as FP and TP", "votes": 1 }, { "id": 1227113, "postDate": "2021-03-05T08:09:08.293Z", "content": "

FYI - see this discussion for requirements for hand labeled / re-annotated train images or external images
\nhttps://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/222644

", "rawMarkdown": "FYI - see this discussion for requirements for hand labeled / re-annotated train images or external images\nhttps://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/222644", "votes": 1 }, { "id": 1227053, "postDate": "2021-03-05T06:18:55.453Z", "content": "
from common import *\nfrom ranzcr import *\n\ndata_dir = '/root/share1/kaggle/2021/ranzcr/data/ranzcr-clip-catheter-line-classification'\n\ndf_annotation = pd.read_csv(data_dir + '/train_annotations.more.fix.csv')\ndf = pd.read_csv(data_dir + '/train.more.fix.csv')\ndf = df[df.fold!=1]\n\n#-------------------------\nclahe = cv2.createCLAHE(clipLimit=8, tileGridSize=(8, 8))\n#clahe = cv2.createCLAHE(clipLimit=13.5, tileGridSize=(8, 8))\n\n\n# is_a=1  #ground truth\n# dump_dir = '/root/share1/kaggle/2021/ranzcr/data/ranzcr-clip-catheter-line-classification/hand-label/truth'\n# os.makedirs(dump_dir, exist_ok=True)\n# df = df[df.is_annotate==1]\n\nis_a=0\ndump_dir = '/root/share1/kaggle/2021/ranzcr/data/ranzcr-clip-catheter-line-classification/hand-label/try'\nos.makedirs(dump_dir, exist_ok=True)\ndf = df[df.is_annotate==0]\n\n\n#-----------------------------------------\ndf = df.reset_index(drop=True)\nindex = np.arange(len(df))\nfor i in index[:300]:\n    print(i)\n    d = df.iloc[i]\n\n    image = cv2.imread(data_dir+'/train/%s.jpg'%d.StudyInstanceUID, cv2.IMREAD_GRAYSCALE)\n    image = cv2.resize(image, dsize=(768,768), interpolation=cv2.INTER_LINEAR)\n    image = 255-image\n    image = clahe.apply(image)\n\n    is_cvc=0\n    if d[target_col[7]]==1:\n        is_cvc=1\n        draw_shadow_text(image, target_col[7], (5,20), 0.6, 255,1)\n\n    if d[target_col[8]] == 1:\n        is_cvc=1\n        draw_shadow_text(image, target_col[8], (5,40), 0.6, 255, 1)\n\n    if d[target_col[9]] == 1:\n        is_cvc=1\n        draw_shadow_text(image, target_col[9], (5,60), 0.6, 255, 1)\n    if is_cvc == 0 : continue\n\n    if is_a==1:\n        df_a = df_annotation[df_annotation.StudyInstanceUID==d.StudyInstanceUID]\n        for row in df_a.iterrows():\n            if 'CVC' in row[1].label:\n                point = np.array(eval(row[1].data))*[[768,768]]\n                point = point.astype(np.int32)\n\n                x,y = point[0]\n                cv2.circle(image,(x,y),32,255,2,cv2.LINE_AA)\n                x,y = point[-1]\n                cv2.circle(image,(x,y),32,255,2,cv2.LINE_AA)\n\n                for x,y in point:\n                    cv2.circle(image,(x,y),1,255,-1,cv2.LINE_AA)\n\n\n    image_show('image', image)\n    cv2.waitKey(1)\n    cv2.imwrite(dump_dir+'/%s.png'%d.StudyInstanceUID,image)\n
", "rawMarkdown": "```\nfrom common import *\nfrom ranzcr import *\n\ndata_dir = '/root/share1/kaggle/2021/ranzcr/data/ranzcr-clip-catheter-line-classification'\n\ndf_annotation = pd.read_csv(data_dir + '/train_annotations.more.fix.csv')\ndf = pd.read_csv(data_dir + '/train.more.fix.csv')\ndf = df[df.fold!=1]\n \n#-------------------------\nclahe = cv2.createCLAHE(clipLimit=8, tileGridSize=(8, 8))\n#clahe = cv2.createCLAHE(clipLimit=13.5, tileGridSize=(8, 8))\n \n\n# is_a=1 #ground truth\n# dump_dir = '/root/share1/kaggle/2021/ranzcr/data/ranzcr-clip-catheter-line-classification/hand-label/truth'\n# os.makedirs(dump_dir, exist_ok=True)\n# df = df[df.is_annotate==1]\n \nis_a=0\ndump_dir = '/root/share1/kaggle/2021/ranzcr/data/ranzcr-clip-catheter-line-classification/hand-label/try'\nos.makedirs(dump_dir, exist_ok=True)\ndf = df[df.is_annotate==0]\n\n\n#-----------------------------------------\ndf = df.reset_index(drop=True)\nindex = np.arange(len(df))\nfor i in index[:300]:\n print(i)\n d = df.iloc[i]\n\n image = cv2.imread(data_dir+'/train/%s.jpg'%d.StudyInstanceUID, cv2.IMREAD_GRAYSCALE)\n image = cv2.resize(image, dsize=(768,768), interpolation=cv2.INTER_LINEAR)\n image = 255-image\n image = clahe.apply(image)\n\n is_cvc=0\n if d[target_col[7]]==1:\n is_cvc=1\n draw_shadow_text(image, target_col[7], (5,20), 0.6, 255,1)\n\n if d[target_col[8]] == 1:\n is_cvc=1\n draw_shadow_text(image, target_col[8], (5,40), 0.6, 255, 1)\n\n if d[target_col[9]] == 1:\n is_cvc=1\n draw_shadow_text(image, target_col[9], (5,60), 0.6, 255, 1)\n if is_cvc == 0 : continue\n\n if is_a==1:\n df_a = df_annotation[df_annotation.StudyInstanceUID==d.StudyInstanceUID]\n for row in df_a.iterrows():\n if 'CVC' in row[1].label:\n point = np.array(eval(row[1].data))*[[768,768]]\n point = point.astype(np.int32)\n\n x,y = point[0]\n cv2.circle(image,(x,y),32,255,2,cv2.LINE_AA)\n x,y = point[-1]\n cv2.circle(image,(x,y),32,255,2,cv2.LINE_AA)\n\n for x,y in point:\n cv2.circle(image,(x,y),1,255,-1,cv2.LINE_AA)\n\n\n image_show('image', image)\n cv2.waitKey(1)\n cv2.imwrite(dump_dir+'/%s.png'%d.StudyInstanceUID,image)\n\n\n```", "votes": 1 }, { "id": 1234764, "postDate": "2021-03-11T15:06:23.870Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1227209, "author_name": "hengck23", "author_url": "", "post_date": "2021-03-05T09:55:35.497000", "content": "

Alternative approach is to use RFCx method:
\n-use model to generate high confidence candidates
\n-then use hand label to mark as FP and TP

", "votes": 1, "replies": [] }, { "id": 1227113, "author_name": "something4kag", "author_url": "", "post_date": "2021-03-05T08:09:08.293000", "content": "

FYI - see this discussion for requirements for hand labeled / re-annotated train images or external images
\nhttps://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/222644

", "votes": 1, "replies": [] }, { "id": 1227053, "author_name": "hengck23", "author_url": "", "post_date": "2021-03-05T06:18:55.453000", "content": "
from common import *\nfrom ranzcr import *\n\ndata_dir = '/root/share1/kaggle/2021/ranzcr/data/ranzcr-clip-catheter-line-classification'\n\ndf_annotation = pd.read_csv(data_dir + '/train_annotations.more.fix.csv')\ndf = pd.read_csv(data_dir + '/train.more.fix.csv')\ndf = df[df.fold!=1]\n\n#-------------------------\nclahe = cv2.createCLAHE(clipLimit=8, tileGridSize=(8, 8))\n#clahe = cv2.createCLAHE(clipLimit=13.5, tileGridSize=(8, 8))\n\n\n# is_a=1  #ground truth\n# dump_dir = '/root/share1/kaggle/2021/ranzcr/data/ranzcr-clip-catheter-line-classification/hand-label/truth'\n# os.makedirs(dump_dir, exist_ok=True)\n# df = df[df.is_annotate==1]\n\nis_a=0\ndump_dir = '/root/share1/kaggle/2021/ranzcr/data/ranzcr-clip-catheter-line-classification/hand-label/try'\nos.makedirs(dump_dir, exist_ok=True)\ndf = df[df.is_annotate==0]\n\n\n#-----------------------------------------\ndf = df.reset_index(drop=True)\nindex = np.arange(len(df))\nfor i in index[:300]:\n    print(i)\n    d = df.iloc[i]\n\n    image = cv2.imread(data_dir+'/train/%s.jpg'%d.StudyInstanceUID, cv2.IMREAD_GRAYSCALE)\n    image = cv2.resize(image, dsize=(768,768), interpolation=cv2.INTER_LINEAR)\n    image = 255-image\n    image = clahe.apply(image)\n\n    is_cvc=0\n    if d[target_col[7]]==1:\n        is_cvc=1\n        draw_shadow_text(image, target_col[7], (5,20), 0.6, 255,1)\n\n    if d[target_col[8]] == 1:\n        is_cvc=1\n        draw_shadow_text(image, target_col[8], (5,40), 0.6, 255, 1)\n\n    if d[target_col[9]] == 1:\n        is_cvc=1\n        draw_shadow_text(image, target_col[9], (5,60), 0.6, 255, 1)\n    if is_cvc == 0 : continue\n\n    if is_a==1:\n        df_a = df_annotation[df_annotation.StudyInstanceUID==d.StudyInstanceUID]\n        for row in df_a.iterrows():\n            if 'CVC' in row[1].label:\n                point = np.array(eval(row[1].data))*[[768,768]]\n                point = point.astype(np.int32)\n\n                x,y = point[0]\n                cv2.circle(image,(x,y),32,255,2,cv2.LINE_AA)\n                x,y = point[-1]\n                cv2.circle(image,(x,y),32,255,2,cv2.LINE_AA)\n\n                for x,y in point:\n                    cv2.circle(image,(x,y),1,255,-1,cv2.LINE_AA)\n\n\n    image_show('image', image)\n    cv2.waitKey(1)\n    cv2.imwrite(dump_dir+'/%s.png'%d.StudyInstanceUID,image)\n
", "votes": 1, "replies": [] }, { "id": 1234764, "author_name": "", "author_url": "", "post_date": "2021-03-11T15:06:23.870000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1227050": "i just label difficult cvc cases (endpoint only)\n\n![](https://i.ibb.co/jb5vy1X/Selection-047.png)\n![](https://i.ibb.co/9YX7Y0d/Selection-046.png)\n![](https://i.ibb.co/VY6QJ6n/Selection-048.png)", "1227209": "Alternative approach is to use RFCx method:\n-use model to generate high confidence candidates\n-then use hand label to mark as FP and TP", "1227113": "FYI - see this discussion for requirements for hand labeled / re-annotated train images or external images\nhttps://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/222644", "1227053": "```\nfrom common import *\nfrom ranzcr import *\n\ndata_dir = '/root/share1/kaggle/2021/ranzcr/data/ranzcr-clip-catheter-line-classification'\n\ndf_annotation = pd.read_csv(data_dir + '/train_annotations.more.fix.csv')\ndf = pd.read_csv(data_dir + '/train.more.fix.csv')\ndf = df[df.fold!=1]\n \n#-------------------------\nclahe = cv2.createCLAHE(clipLimit=8, tileGridSize=(8, 8))\n#clahe = cv2.createCLAHE(clipLimit=13.5, tileGridSize=(8, 8))\n \n\n# is_a=1 #ground truth\n# dump_dir = '/root/share1/kaggle/2021/ranzcr/data/ranzcr-clip-catheter-line-classification/hand-label/truth'\n# os.makedirs(dump_dir, exist_ok=True)\n# df = df[df.is_annotate==1]\n \nis_a=0\ndump_dir = '/root/share1/kaggle/2021/ranzcr/data/ranzcr-clip-catheter-line-classification/hand-label/try'\nos.makedirs(dump_dir, exist_ok=True)\ndf = df[df.is_annotate==0]\n\n\n#-----------------------------------------\ndf = df.reset_index(drop=True)\nindex = np.arange(len(df))\nfor i in index[:300]:\n print(i)\n d = df.iloc[i]\n\n image = cv2.imread(data_dir+'/train/%s.jpg'%d.StudyInstanceUID, cv2.IMREAD_GRAYSCALE)\n image = cv2.resize(image, dsize=(768,768), interpolation=cv2.INTER_LINEAR)\n image = 255-image\n image = clahe.apply(image)\n\n is_cvc=0\n if d[target_col[7]]==1:\n is_cvc=1\n draw_shadow_text(image, target_col[7], (5,20), 0.6, 255,1)\n\n if d[target_col[8]] == 1:\n is_cvc=1\n draw_shadow_text(image, target_col[8], (5,40), 0.6, 255, 1)\n\n if d[target_col[9]] == 1:\n is_cvc=1\n draw_shadow_text(image, target_col[9], (5,60), 0.6, 255, 1)\n if is_cvc == 0 : continue\n\n if is_a==1:\n df_a = df_annotation[df_annotation.StudyInstanceUID==d.StudyInstanceUID]\n for row in df_a.iterrows():\n if 'CVC' in row[1].label:\n point = np.array(eval(row[1].data))*[[768,768]]\n point = point.astype(np.int32)\n\n x,y = point[0]\n cv2.circle(image,(x,y),32,255,2,cv2.LINE_AA)\n x,y = point[-1]\n cv2.circle(image,(x,y),32,255,2,cv2.LINE_AA)\n\n for x,y in point:\n cv2.circle(image,(x,y),1,255,-1,cv2.LINE_AA)\n\n\n image_show('image', image)\n cv2.waitKey(1)\n cv2.imwrite(dump_dir+'/%s.png'%d.StudyInstanceUID,image)\n\n\n```", "1234764": "" } }