{ "id": 490895, "title": "Top Solutions From the Previous Competition (BirdCLEF 2023)", "url": "/competitions/birdclef-2024/discussion/490895", "author_name": "Sinan Calisir", "post_date": "2024-04-03T20:47:16.872000", "votes": 38, "comment_count": 6, "views": 0, "content": "

Hi everyone,

\n

Happy to see another BirdCLEF Competition! Here is a list of high scoring solutions from the previous competition that can be helpful to start with!

\n", "messages": [ { "id": 2733890, "postDate": "2024-04-03T20:47:16.873Z", "content": "

Hi everyone,

\n

Happy to see another BirdCLEF Competition! Here is a list of high scoring solutions from the previous competition that can be helpful to start with!

\n", "rawMarkdown": "Hi everyone,\n\nHappy to see another BirdCLEF Competition! Here is a list of high scoring solutions from the previous competition that can be helpful to start with!\n\n- [1st place solution: Correct Data is All You Need](https://www.kaggle.com/competitions/birdclef-2023/discussion/412808)\n- [2nd place solution(7 models ensemble)](https://www.kaggle.com/competitions/birdclef-2023/discussion/412707)\n- [4th Place Solution: Knowledge Distillation is all you need](https://www.kaggle.com/competitions/birdclef-2023/discussion/412753)\n- [5th place Solution](https://www.kaggle.com/competitions/birdclef-2023/discussion/412903)\n- [6th place solution: BirdNET embedding + CNN](https://www.kaggle.com/competitions/birdclef-2023/discussion/412708)\n- [7th place solution](https://www.kaggle.com/competitions/birdclef-2023/discussion/412922)\n- [8th Place Solution: Implementing Multimodal Data Augmentation Methods with YOLOv8, Recommendation Systems, among others](https://www.kaggle.com/competitions/birdclef-2023/discussion/412871)\n- [9th Place Solution: 7 CNN Models Ensemble](https://www.kaggle.com/competitions/birdclef-2023/discussion/412794)\n- [10th place solution with the help of ChatGPT](https://www.kaggle.com/competitions/birdclef-2023/discussion/412713)\n- [12th place solution: 8 CNN models ensemble with OpenVINO](https://www.kaggle.com/competitions/birdclef-2023/discussion/412768)\n- [20th place solution: SED + CNN ensemble using onnx](https://www.kaggle.com/competitions/birdclef-2023/discussion/412742)\n- [24th place solution - pre-training & single model (5 folds ensemble with ONNX)](https://www.kaggle.com/competitions/birdclef-2023/discussion/412996)\n- [37th place solution - TF CNN + BirdNet emb. cls. & XGB](https://www.kaggle.com/competitions/birdclef-2023/discussion/412869)", "votes": 36 }, { "id": 2771413, "postDate": "2024-04-24T08:27:43.760Z", "content": "

Thanks for collecting this list of past solutions.
\nHave you noticed a general trend of what worked back then?

", "rawMarkdown": "Thanks for collecting this list of past solutions.\nHave you noticed a general trend of what worked back then?", "votes": 1, "replies": [ { "id": 2771923, "postDate": "2024-04-24T13:03:08.610Z", "content": "

I would say, in the following order:

\n", "rawMarkdown": "I would say, in the following order:\n\n- Pretraining: Using data from the previous competitions.\n- Augmentations: Different type of augmentations both on audio and specs.\n- Diverse ensemble: Because of the time limit no crazy ensembles but diversity was important.", "votes": 3, "replies": [ { "id": 2772067, "postDate": "2024-04-24T14:11:19.423Z", "content": "

Good insights, thanks.
\nIndeed, it seems that pretraining on previous data (while being careful) is one way to boost performance.

", "rawMarkdown": "Good insights, thanks.\nIndeed, it seems that pretraining on previous data (while being careful) is one way to boost performance.", "votes": 3 }, { "id": 2772072, "postDate": "2024-04-24T14:13:39.927Z", "content": "

It also seems that SED models with various backbones are a good starting point.

", "rawMarkdown": "It also seems that [SED](https://paperswithcode.com/task/sound-event-detection) models with various backbones are a good starting point.", "votes": 2 } ] } ] }, { "id": 2737595, "postDate": "2024-04-05T20:20:05.637Z", "content": "

Very useful! :D

", "rawMarkdown": "Very useful! :D", "votes": 1 }, { "id": 2779207, "postDate": "2024-04-27T15:23:44.640Z", "content": "

Thank you. This is All I NEED !

", "rawMarkdown": "Thank you. This is All I NEED ! ", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 2771413, "author_name": "Yassine Alouini", "author_url": "", "post_date": "2024-04-24T08:27:43.760000", "content": "

Thanks for collecting this list of past solutions.
\nHave you noticed a general trend of what worked back then?

", "votes": 1, "replies": [ { "id": 2771923, "author_name": "Sinan Calisir", "author_url": "", "post_date": "2024-04-24T13:03:08.610000", "content": "

I would say, in the following order:

\n", "votes": 3, "replies": [ { "id": 2772067, "author_name": "Yassine Alouini", "author_url": "", "post_date": "2024-04-24T14:11:19.423000", "content": "

Good insights, thanks.
\nIndeed, it seems that pretraining on previous data (while being careful) is one way to boost performance.

", "votes": 3, "replies": [] }, { "id": 2772072, "author_name": "Yassine Alouini", "author_url": "", "post_date": "2024-04-24T14:13:39.927000", "content": "

It also seems that SED models with various backbones are a good starting point.

", "votes": 2, "replies": [] } ] } ] }, { "id": 2737595, "author_name": "Khoa Tran", "author_url": "", "post_date": "2024-04-05T20:20:05.637000", "content": "

Very useful! :D

", "votes": 1, "replies": [] }, { "id": 2779207, "author_name": "", "author_url": "", "post_date": "2024-04-27T15:23:44.640000", "content": "

Thank you. This is All I NEED !

", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "2733890": "Hi everyone,\n\nHappy to see another BirdCLEF Competition! Here is a list of high scoring solutions from the previous competition that can be helpful to start with!\n\n- [1st place solution: Correct Data is All You Need](https://www.kaggle.com/competitions/birdclef-2023/discussion/412808)\n- [2nd place solution(7 models ensemble)](https://www.kaggle.com/competitions/birdclef-2023/discussion/412707)\n- [4th Place Solution: Knowledge Distillation is all you need](https://www.kaggle.com/competitions/birdclef-2023/discussion/412753)\n- [5th place Solution](https://www.kaggle.com/competitions/birdclef-2023/discussion/412903)\n- [6th place solution: BirdNET embedding + CNN](https://www.kaggle.com/competitions/birdclef-2023/discussion/412708)\n- [7th place solution](https://www.kaggle.com/competitions/birdclef-2023/discussion/412922)\n- [8th Place Solution: Implementing Multimodal Data Augmentation Methods with YOLOv8, Recommendation Systems, among others](https://www.kaggle.com/competitions/birdclef-2023/discussion/412871)\n- [9th Place Solution: 7 CNN Models Ensemble](https://www.kaggle.com/competitions/birdclef-2023/discussion/412794)\n- [10th place solution with the help of ChatGPT](https://www.kaggle.com/competitions/birdclef-2023/discussion/412713)\n- [12th place solution: 8 CNN models ensemble with OpenVINO](https://www.kaggle.com/competitions/birdclef-2023/discussion/412768)\n- [20th place solution: SED + CNN ensemble using onnx](https://www.kaggle.com/competitions/birdclef-2023/discussion/412742)\n- [24th place solution - pre-training & single model (5 folds ensemble with ONNX)](https://www.kaggle.com/competitions/birdclef-2023/discussion/412996)\n- [37th place solution - TF CNN + BirdNet emb. cls. & XGB](https://www.kaggle.com/competitions/birdclef-2023/discussion/412869)", "2771413": "Thanks for collecting this list of past solutions.\nHave you noticed a general trend of what worked back then?", "2737595": "Very useful! :D", "2779207": "Thank you. This is All I NEED ! " } }