This was my 7th competition, and the goal was to get a Silver Medal.
\nBut after finished other competition, I joined this competition lately.
\nSo I first studied other competitor’s public solution.
\nDetectron2 @corochann and yolo5 @Awsaf attracted my eye, and just read @Denny Wang achievement.
\nBut the CV score[mAP@0.5:0.95] didn’t break 1.6, public lb can’t reach to 1.8.
\nWith lucky I implemented Ensemble Strategy with @ZFTurbo’s WBF, and public lb score was reached to 2.72.
\nI tried to implement 5 Folds and TTA, but gave up for my other work.
\nLast Night I ensembled public results with WBF, the results was surprised me to get 2.94(2.82).
\nThis results was enough for me to get Silver Medal.
\nI think there is far distance between GM/Masters and me.
\nI hope to become a Master this year, so I must study hard.
\nThanks to Competition Host, and Congratulations to all winners!
\nSpecial thanks to @corochann @awsaf49 @dennywangdev @zfturbo
This is my submission results.
\n| Model | \nPublic/ Private | \nWBF weight | \nPublic/Private | \n
|---|---|---|---|
| Yolo5 | \n0.243/0.235 | \n[3, 1] | \n0.272/0.256 | \n
| Detectron2 | \n0.221/0.232 | \n\n | \n |
| \n | \n | \n | \n |
| Yolo5 | \n0.243/0.235 | \n[2, 1, 3] | \n0.294/0.282 | \n
| Detectron2 | \n0.221/0.232 | \n\n | \n |
| Public | \n0.246/0.226 | \n\n | \n |
This was my 7th competition, and the goal was to get a Silver Medal.
\nBut after finished other competition, I joined this competition lately.
\nSo I first studied other competitor’s public solution.
\nDetectron2 @corochann and yolo5 @Awsaf attracted my eye, and just read @Denny Wang achievement.
\nBut the CV score[mAP@0.5:0.95] didn’t break 1.6, public lb can’t reach to 1.8.
\nWith lucky I implemented Ensemble Strategy with @ZFTurbo’s WBF, and public lb score was reached to 2.72.
\nI tried to implement 5 Folds and TTA, but gave up for my other work.
\nLast Night I ensembled public results with WBF, the results was surprised me to get 2.94(2.82).
\nThis results was enough for me to get Silver Medal.
\nI think there is far distance between GM/Masters and me.
\nI hope to become a Master this year, so I must study hard.
\nThanks to Competition Host, and Congratulations to all winners!
\nSpecial thanks to @corochann @awsaf49 @dennywangdev @zfturbo
This is my submission results.
\n| Model | \nPublic/ Private | \nWBF weight | \nPublic/Private | \n
|---|---|---|---|
| Yolo5 | \n0.243/0.235 | \n[3, 1] | \n0.272/0.256 | \n
| Detectron2 | \n0.221/0.232 | \n\n | \n |
| \n | \n | \n | \n |
| Yolo5 | \n0.243/0.235 | \n[2, 1, 3] | \n0.294/0.282 | \n
| Detectron2 | \n0.221/0.232 | \n\n | \n |
| Public | \n0.246/0.226 | \n\n | \n |
Congrats on silver medal @xiaojin712 and thanks for sharing summarize solution
", "rawMarkdown": "Congrats on silver medal @xiaojin712 and thanks for sharing summarize solution", "votes": 3, "replies": [ { "id": 1257701, "postDate": "2021-03-31T03:37:25.893Z", "content": "Thank you , and Congrats on your Silver Medal too :).
", "rawMarkdown": "Thank you , and Congrats on your Silver Medal too :).", "votes": 1 } ] }, { "id": 1257652, "postDate": "2021-03-31T02:40:31.873Z", "content": "Congratulations to get the silver medal.
", "rawMarkdown": "Congratulations to get the silver medal.", "votes": 1, "replies": [ { "id": 1257832, "postDate": "2021-03-31T06:13:04.450Z", "content": "Thanks a lot :)
", "rawMarkdown": "Thanks a lot :)", "votes": 1 } ] }, { "id": 1257927, "postDate": "2021-03-31T08:12:03.470Z", "content": "Congratz 😀
", "rawMarkdown": "Congratz 😀", "votes": 2, "replies": [ { "id": 1257951, "postDate": "2021-03-31T08:28:36.250Z", "content": "Thanks a lot 👍, And Congrats on your Medal too 😃.
\nYour kernel helped me a lot.
Hi thank you, can I ask how can you choose a suitable WBF weight like [3,1]?
", "rawMarkdown": "Hi thank you, can I ask how can you choose a suitable WBF weight like [3,1]?", "votes": 2, "replies": [ { "id": 1257871, "postDate": "2021-03-31T06:58:10.167Z", "content": "\n\noutput = pd.DataFrame(columns = [\"image_id\", \"PredictionString\"])
\n
\n for image_id in tqdm(image_ids):
\n boxes_list = []
\n scores_list = []
\n labels_list = []
\n # For each model, get PredictionString of image_id as a dict
\n boxes, scores, labels = getitem(sub_yolo, image_id)
\n boxes_list.append(boxes)
\n scores_list.append(scores)
\n labels_list.append(labels)
\n boxes, scores, labels = getitem(sub1, image_id)
\n boxes_list.append(boxes)
\n scores_list.append(scores)
\n labels_list.append(labels)
\n boxes, scores, labels = getitem(sub2, image_id)
\n boxes_list.append(boxes)
\n scores_list.append(scores)
\n labels_list.append(labels)
\n # weights = [1, 4] #0.279
\n # weights = [1, 3.5] #0.285
\n # weights = [1, 3] #0.286
\n # weights = [1, 2.5] #0.270
\n # weights = [1, 2] #0.272
\n # weights = [1, 1] #0.241
\n # weights = [1, 0] #0.21
\n # weights = [0, 1] #0.247
\n # weights = [2, 1, 3] #0.294
\n # weights = [4, 1, 3] #0.289
\n # weights = [3, 1] #0.272
\n weights = [3, 1, 2] #0.291
\n boxes, scores, labels = weighted_boxes_fusion(boxes_list, scores_list, labels_list, weights, 0.4, 0.0001, )
This is my code for WBF ensemble, Finally I chose with best public score.
", "rawMarkdown": "> output = pd.DataFrame(columns = [\"image_id\", \"PredictionString\"]) \n for image_id in tqdm(image_ids):\n boxes_list = []\n scores_list = []\n labels_list = []\n # For each model, get PredictionString of image_id as a dict\n boxes, scores, labels = getitem(sub_yolo, image_id)\n boxes_list.append(boxes)\n scores_list.append(scores)\n labels_list.append(labels)\n boxes, scores, labels = getitem(sub1, image_id)\n boxes_list.append(boxes)\n scores_list.append(scores)\n labels_list.append(labels)\n boxes, scores, labels = getitem(sub2, image_id)\n boxes_list.append(boxes)\n scores_list.append(scores)\n labels_list.append(labels) \n # weights = [1, 4] #0.279\n # weights = [1, 3.5] #0.285\n # weights = [1, 3] #0.286\n # weights = [1, 2.5] #0.270\n # weights = [1, 2] #0.272\n # weights = [1, 1] #0.241\n # weights = [1, 0] #0.21\n # weights = [0, 1] #0.247 \n # weights = [2, 1, 3] #0.294\n # weights = [4, 1, 3] #0.289\n # weights = [3, 1] #0.272\n weights = [3, 1, 2] #0.291 \n boxes, scores, labels = weighted_boxes_fusion(boxes_list, scores_list, labels_list, weights, 0.4, 0.0001, )\n\nThis is my code for WBF ensemble, Finally I chose with best public score.", "votes": 1 } ] }, { "id": 1257674, "postDate": "2021-03-31T03:03:37.067Z", "content": "Congrats and thanks for sharing!
", "rawMarkdown": "Congrats and thanks for sharing!", "votes": 2, "replies": [ { "id": 1257678, "postDate": "2021-03-31T03:11:55.913Z", "content": "Thank you , and Congrats on your Medal too :).
", "rawMarkdown": "Thank you , and Congrats on your Medal too :).", "votes": 2 } ] }, { "id": 1257654, "postDate": "2021-03-31T02:41:05.560Z", "content": "@xiaojin712 Congratulations on Silver Finish and Thanks for sharing your approach
", "rawMarkdown": "@xiaojin712 Congratulations on Silver Finish and Thanks for sharing your approach", "votes": 2, "replies": [ { "id": 1257657, "postDate": "2021-03-31T02:45:40.030Z", "content": "Thank you :)
", "rawMarkdown": "Thank you :)", "votes": 1 } ] } ], "comments": [ { "id": 1257692, "author_name": "KhanhVD", "author_url": "", "post_date": "2021-03-31T03:24:14.547000", "content": "Congrats on silver medal @xiaojin712 and thanks for sharing summarize solution
", "votes": 3, "replies": [ { "id": 1257701, "author_name": "Wang Xing", "author_url": "", "post_date": "2021-03-31T03:37:25.893000", "content": "Thank you , and Congrats on your Silver Medal too :).
", "votes": 1, "replies": [] } ] }, { "id": 1257652, "author_name": "TianLong Ye", "author_url": "", "post_date": "2021-03-31T02:40:31.873000", "content": "Congratulations to get the silver medal.
", "votes": 1, "replies": [ { "id": 1257832, "author_name": "Wang Xing", "author_url": "", "post_date": "2021-03-31T06:13:04.450000", "content": "Thanks a lot :)
", "votes": 1, "replies": [] } ] }, { "id": 1257927, "author_name": "Awsaf", "author_url": "", "post_date": "2021-03-31T08:12:03.470000", "content": "Congratz 😀
", "votes": 2, "replies": [ { "id": 1257951, "author_name": "Wang Xing", "author_url": "", "post_date": "2021-03-31T08:28:36.250000", "content": "Thanks a lot 👍, And Congrats on your Medal too 😃.
\nYour kernel helped me a lot.
Hi thank you, can I ask how can you choose a suitable WBF weight like [3,1]?
", "votes": 2, "replies": [ { "id": 1257871, "author_name": "Wang Xing", "author_url": "", "post_date": "2021-03-31T06:58:10.167000", "content": "\n\noutput = pd.DataFrame(columns = [\"image_id\", \"PredictionString\"])
\n
\n for image_id in tqdm(image_ids):
\n boxes_list = []
\n scores_list = []
\n labels_list = []
\n # For each model, get PredictionString of image_id as a dict
\n boxes, scores, labels = getitem(sub_yolo, image_id)
\n boxes_list.append(boxes)
\n scores_list.append(scores)
\n labels_list.append(labels)
\n boxes, scores, labels = getitem(sub1, image_id)
\n boxes_list.append(boxes)
\n scores_list.append(scores)
\n labels_list.append(labels)
\n boxes, scores, labels = getitem(sub2, image_id)
\n boxes_list.append(boxes)
\n scores_list.append(scores)
\n labels_list.append(labels)
\n # weights = [1, 4] #0.279
\n # weights = [1, 3.5] #0.285
\n # weights = [1, 3] #0.286
\n # weights = [1, 2.5] #0.270
\n # weights = [1, 2] #0.272
\n # weights = [1, 1] #0.241
\n # weights = [1, 0] #0.21
\n # weights = [0, 1] #0.247
\n # weights = [2, 1, 3] #0.294
\n # weights = [4, 1, 3] #0.289
\n # weights = [3, 1] #0.272
\n weights = [3, 1, 2] #0.291
\n boxes, scores, labels = weighted_boxes_fusion(boxes_list, scores_list, labels_list, weights, 0.4, 0.0001, )
This is my code for WBF ensemble, Finally I chose with best public score.
", "votes": 1, "replies": [] } ] }, { "id": 1257674, "author_name": "Wonho Song", "author_url": "", "post_date": "2021-03-31T03:03:37.067000", "content": "Congrats and thanks for sharing!
", "votes": 2, "replies": [ { "id": 1257678, "author_name": "Wang Xing", "author_url": "", "post_date": "2021-03-31T03:11:55.913000", "content": "Thank you , and Congrats on your Medal too :).
", "votes": 2, "replies": [] } ] }, { "id": 1257654, "author_name": "Tensor Girl", "author_url": "", "post_date": "2021-03-31T02:41:05.560000", "content": "@xiaojin712 Congratulations on Silver Finish and Thanks for sharing your approach
", "votes": 2, "replies": [ { "id": 1257657, "author_name": "Wang Xing", "author_url": "", "post_date": "2021-03-31T02:45:40.030000", "content": "Thank you :)
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "1257648": "This was my 7th competition, and the goal was to get a Silver Medal.\nBut after finished other competition, I joined this competition lately.\nSo I first studied other competitor’s public solution.\nDetectron2 [@corochann](https://www.kaggle.com/corochann/vinbigdata-detectron2-train) and yolo5 [@Awsaf](https://www.kaggle.com/awsaf49/vinbigdata-cxr-ad-yolov5-14-class-train) attracted my eye, and just read [@Denny Wang](https://www.kaggle.com/dennywangdev) [achievement](https://www.kaggle.com/c/vinbigdata-chest-xray-abnormalities-detection/discussion/219925#1220224).\nBut the CV score[mAP@0.5:0.95] didn’t break 1.6, public lb can’t reach to 1.8.\nWith lucky I implemented Ensemble Strategy with [@ZFTurbo](https://www.kaggle.com/zfturbo)’s WBF, and public lb score was reached to 2.72.\nI tried to implement 5 Folds and TTA, but gave up for my other work.\nLast Night I ensembled public results with WBF, the results was surprised me to get 2.94(2.82).\nThis results was enough for me to get Silver Medal.\nI think there is far distance between GM/Masters and me.\nI hope to become a Master this year, so I must **study** hard.\nThanks to Competition Host, and Congratulations to all winners!\nSpecial thanks to @corochann @awsaf49 @dennywangdev @zfturbo \n\nThis is my submission results.\n|Model\t|Public/ Private\t|WBF weight\t|Public/Private|\n| --- | --- |\n|Yolo5\t|0.243/0.235\t|[3, 1]\t|0.272/0.256|\n|Detectron2\t|0.221/0.232|\t\t||\n|\t\t\t| ||\n|Yolo5\t|0.243/0.235\t|[2, 1, 3]\t|**0.294/0.282**|\n|Detectron2\t|0.221/0.232| ||\t\t\n|Public\t|0.246/0.226| ||\t\t\n", "1257692": "Congrats on silver medal @xiaojin712 and thanks for sharing summarize solution", "1257652": "Congratulations to get the silver medal.", "1257927": "Congratz 😀", "1257860": "Hi thank you, can I ask how can you choose a suitable WBF weight like [3,1]?", "1257674": "Congrats and thanks for sharing!", "1257654": "@xiaojin712 Congratulations on Silver Finish and Thanks for sharing your approach" } }