\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4931964%2Fe83af1b1abef9051fceb33e93a4f0eca%2F__results___61_0.png?generation=1596705036976428&alt=media\")
Thanks for all the work done by the contributor of the notebooks community, thanks for their sharing, we really learned a lot from it, and we thanks for all the participant's work that makes the leaderboard more energetic!
\nIs not easy to get 340th place, we try to use a number of methods but hard to improve.
\nThrough our tireless efforts, we got 0.7666 on public and 0.6916 on private.
\nThese are some strategies for our final used submission.
Here is our dataset of cleaning markings. link to pure-box Thanks to @songzhijin @Abusi Brown They are my team member, thanks for their great effort and amazing work on the data cleaning.
\nWe used this tool to relabel the box. Labelimg
\nAnd I code this notebook for some format converting.Train Marking(csv) to PascalVOC(xml)
We used 1024x1024 images, Stratified 5 Folds Splitting, some Augmentation like others, on Efficientdet-D5.
\nAnd we used apex.amp so that we can use batch-size=2 in 1024 to have a better loss on training.
We used 2 times of Pseudo Labeling, in the PLabeling we choose WBF iou_thr=0.2, skip_box_thr=0.33, each of them got training 3 epochs, the data consist of training data with our cleaned label and testing data with PLabel .
\nIn the final prediction, we used back to iou_thr=0.44, skip_box_thr=0.43 of WBF to get a more precise result.
The detection effect like this:
\n
Here is our notebook link. Efficientdet-with-double-Pseudo-Labeling
\nThanks for your reading, I hope we can give you some inspiration through this page.
And I want to say in the end, thanks a lot to kaggle team, we really appreciate they managed to contain the rules of the competition and gave us the platform to communicate with others!
", "messages": [ { "id": 960246, "postDate": "2020-08-06T09:04:42.117Z", "content": "Thanks for all the work done by the contributor of the notebooks community, thanks for their sharing, we really learned a lot from it, and we thanks for all the participant's work that makes the leaderboard more energetic!
\nIs not easy to get 340th place, we try to use a number of methods but hard to improve.
\nThrough our tireless efforts, we got 0.7666 on public and 0.6916 on private.
\nThese are some strategies for our final used submission.
Here is our dataset of cleaning markings. link to pure-box Thanks to @songzhijin @Abusi Brown They are my team member, thanks for their great effort and amazing work on the data cleaning.
\nWe used this tool to relabel the box. Labelimg
\nAnd I code this notebook for some format converting.Train Marking(csv) to PascalVOC(xml)
We used 1024x1024 images, Stratified 5 Folds Splitting, some Augmentation like others, on Efficientdet-D5.
\nAnd we used apex.amp so that we can use batch-size=2 in 1024 to have a better loss on training.
We used 2 times of Pseudo Labeling, in the PLabeling we choose WBF iou_thr=0.2, skip_box_thr=0.33, each of them got training 3 epochs, the data consist of training data with our cleaned label and testing data with PLabel .
\nIn the final prediction, we used back to iou_thr=0.44, skip_box_thr=0.43 of WBF to get a more precise result.
The detection effect like this:
\n
Here is our notebook link. Efficientdet-with-double-Pseudo-Labeling
\nThanks for your reading, I hope we can give you some inspiration through this page.
And I want to say in the end, thanks a lot to kaggle team, we really appreciate they managed to contain the rules of the competition and gave us the platform to communicate with others!
", "rawMarkdown": "Thanks for all the work done by the contributor of the notebooks community, thanks for their sharing, we really learned a lot from it, and we thanks for all the participant's work that makes the leaderboard more energetic!\n\nIs not easy to get 340th place, we try to use a number of methods but hard to improve. \nThrough our tireless efforts, we got 0.7666 on public and 0.6916 on private.\nThese are some strategies for our final used submission.\n\n### Data Cleaning ###\n* We deleted some boxes that squared wheat which cannot be seen very clearly and some boxes that appeared in the edge of the picture while squared nothing.\n* We also resize the boxes that are squared awn of wheat to only the wheat head. We did not resize all pictures because we concerned about the reduction of the predicted box may lead to a low percentage, for the IOU of predicted and ground truth > 0.5 got hit.\n\nHere is our dataset of cleaning markings. [link to pure-box](https://www.kaggle.com/dpyrtfq2372/pure-box) Thanks to [@songzhijin](https://www.kaggle.com/songzhijin) [@Abusi Brown](https://www.kaggle.com/luluqie) They are my team member, thanks for their great effort and amazing work on the data cleaning. \nWe used this tool to relabel the box. [Labelimg](https://github.com/tzutalin/labelImg)\nAnd I code this notebook for some format converting.[Train Marking(csv) to PascalVOC(xml)](https://www.kaggle.com/dpyrtfq2372/train-marking-to-pascalvoc/data)\n\n### Training ###\nWe used 1024x1024 images, Stratified 5 Folds Splitting, some Augmentation like others, on Efficientdet-D5.\nAnd we used apex.amp so that we can use batch-size=2 in 1024 to have a better loss on training.\n\n### Inferencing ###\nWe used 2 times of Pseudo Labeling, in the PLabeling we choose WBF iou\\_thr=0.2, skip\\_box\\_thr=0.33, each of them got training 3 epochs, the data consist of training data with our cleaned label and testing data with PLabel .\nIn the final prediction, we used back to iou\\_thr=0.44, skip\\_box\\_thr=0.43 of WBF to get a more precise result.\n\nThe detection effect like this:\n\n\n\nHere is our notebook link. [Efficientdet-with-double-Pseudo-Labeling](https://www.kaggle.com/dpyrtfq2372/efficientdet-with-double-psudo-labeling)\nThanks for your reading, I hope we can give you some inspiration through this page.\n\nAnd I want to say in the end, thanks a lot to kaggle team, we really appreciate they managed to contain the rules of the competition and gave us the platform to communicate with others!", "votes": 7 }, { "id": 960558, "postDate": "2020-08-06T14:03:41.080Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 960678, "postDate": "2020-08-06T15:37:35.620Z", "content": "Thank you😄
", "rawMarkdown": "Thank you😄 " } ] } ], "comments": [ { "id": 960558, "author_name": "", "author_url": "", "post_date": "2020-08-06T14:03:41.080000", "content": "", "votes": 0, "replies": [ { "id": 960678, "author_name": "roc", "author_url": "", "post_date": "2020-08-06T15:37:35.620000", "content": "Thank you😄
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "960246": "Thanks for all the work done by the contributor of the notebooks community, thanks for their sharing, we really learned a lot from it, and we thanks for all the participant's work that makes the leaderboard more energetic!\n\nIs not easy to get 340th place, we try to use a number of methods but hard to improve. \nThrough our tireless efforts, we got 0.7666 on public and 0.6916 on private.\nThese are some strategies for our final used submission.\n\n### Data Cleaning ###\n* We deleted some boxes that squared wheat which cannot be seen very clearly and some boxes that appeared in the edge of the picture while squared nothing.\n* We also resize the boxes that are squared awn of wheat to only the wheat head. We did not resize all pictures because we concerned about the reduction of the predicted box may lead to a low percentage, for the IOU of predicted and ground truth > 0.5 got hit.\n\nHere is our dataset of cleaning markings. [link to pure-box](https://www.kaggle.com/dpyrtfq2372/pure-box) Thanks to [@songzhijin](https://www.kaggle.com/songzhijin) [@Abusi Brown](https://www.kaggle.com/luluqie) They are my team member, thanks for their great effort and amazing work on the data cleaning. \nWe used this tool to relabel the box. [Labelimg](https://github.com/tzutalin/labelImg)\nAnd I code this notebook for some format converting.[Train Marking(csv) to PascalVOC(xml)](https://www.kaggle.com/dpyrtfq2372/train-marking-to-pascalvoc/data)\n\n### Training ###\nWe used 1024x1024 images, Stratified 5 Folds Splitting, some Augmentation like others, on Efficientdet-D5.\nAnd we used apex.amp so that we can use batch-size=2 in 1024 to have a better loss on training.\n\n### Inferencing ###\nWe used 2 times of Pseudo Labeling, in the PLabeling we choose WBF iou\\_thr=0.2, skip\\_box\\_thr=0.33, each of them got training 3 epochs, the data consist of training data with our cleaned label and testing data with PLabel .\nIn the final prediction, we used back to iou\\_thr=0.44, skip\\_box\\_thr=0.43 of WBF to get a more precise result.\n\nThe detection effect like this:\n\n\n\nHere is our notebook link. [Efficientdet-with-double-Pseudo-Labeling](https://www.kaggle.com/dpyrtfq2372/efficientdet-with-double-psudo-labeling)\nThanks for your reading, I hope we can give you some inspiration through this page.\n\nAnd I want to say in the end, thanks a lot to kaggle team, we really appreciate they managed to contain the rules of the competition and gave us the platform to communicate with others!", "960558": "" } }