{ "id": 415176, "title": "Unet Baseline using PyTorch - [LB - 0.628]", "url": "/competitions/google-research-identify-contrails-reduce-global-warming/discussion/415176", "author_name": "Shashwat Raman", "post_date": "2023-06-05T13:15:41.309000", "votes": 31, "comment_count": 3, "views": 0, "content": "

Unet Baseline using PyTorch - [LB - 0.628]

\n

I've created a Unet baseline for this competition. It achieves a score of 0.580 on the leaderboard.

\n

Training Notebook: Simple Unet Baseline (Train)
\nInference Notebook: Simple Unet Baseline (Infer)
\nDataset Notebook : Contrails Dataset (Ash Color)

\n

To get a better understanding of how everything works, please do look at the notebook used to create the dataset.

\n

Some points:

\n\n

Please upvote if you find this useful. If you don't understand anything, please do ask me, I'm really happy to help :)

\n

Update
\nUsing EfficientNet B3, adding some augmentations and training for 30 epochs, the new version of the baseline gets a lb score of 0.628.

", "messages": [ { "id": 2288554, "postDate": "2023-06-05T13:15:41.310Z", "content": "

Unet Baseline using PyTorch - [LB - 0.628]

\n

I've created a Unet baseline for this competition. It achieves a score of 0.580 on the leaderboard.

\n

Training Notebook: Simple Unet Baseline (Train)
\nInference Notebook: Simple Unet Baseline (Infer)
\nDataset Notebook : Contrails Dataset (Ash Color)

\n

To get a better understanding of how everything works, please do look at the notebook used to create the dataset.

\n

Some points:

\n\n

Please upvote if you find this useful. If you don't understand anything, please do ask me, I'm really happy to help :)

\n

Update
\nUsing EfficientNet B3, adding some augmentations and training for 30 epochs, the new version of the baseline gets a lb score of 0.628.

", "rawMarkdown": "# Unet Baseline using PyTorch - [LB - 0.628]\n\nI've created a Unet baseline for this competition. It achieves a score of 0.580 on the leaderboard.\n\n**Training Notebook**: [Simple Unet Baseline (Train)](https://www.kaggle.com/code/shashwatraman/simple-unet-pytorch-baseline-train)\n**Inference Notebook**: [Simple Unet Baseline (Infer)](https://www.kaggle.com/code/shashwatraman/simple-unet-pytorch-baseline-infer)\n**Dataset Notebook** : [Contrails Dataset (Ash Color)](https://www.kaggle.com/code/shashwatraman/contrails-dataset-ash-color)\n\nTo get a better understanding of how everything works, please do look at the notebook used to create the dataset.\n\nSome points:\n- Library: Smp\n- Data: Ash Color images (With only the labeled frames and human_pixel_masks)\n- Backbone: EfficientNet-B0\n- Postprocessing: Finding the best threshold\n\nPlease upvote if you find this useful. If you don't understand anything, please do ask me, I'm really happy to help :)\n\nUpdate\nUsing EfficientNet B3, adding some augmentations and training for 30 epochs, the new version of the baseline gets a lb score of 0.628.\n\n\n\n", "votes": 28 }, { "id": 2336638, "postDate": "2023-07-09T13:50:35.313Z", "content": "

Update:
\nUsing EfficientNet B3, adding some augmentations and training for 30 epochs, the new version of the baseline gets a lb score of 0.628.

", "rawMarkdown": "Update:\nUsing EfficientNet B3, adding some augmentations and training for 30 epochs, the new version of the baseline gets a lb score of 0.628.", "votes": 1 }, { "id": 2292142, "postDate": "2023-06-08T04:45:12.700Z", "content": "

Very clean, nice !👍

", "rawMarkdown": "Very clean, nice !👍", "votes": 1 }, { "id": 2330363, "postDate": "2023-07-04T23:40:42.900Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 2336638, "author_name": "Shashwat Raman", "author_url": "", "post_date": "2023-07-09T13:50:35.313000", "content": "

Update:
\nUsing EfficientNet B3, adding some augmentations and training for 30 epochs, the new version of the baseline gets a lb score of 0.628.

", "votes": 1, "replies": [] }, { "id": 2292142, "author_name": "Thomas", "author_url": "", "post_date": "2023-06-08T04:45:12.700000", "content": "

Very clean, nice !👍

", "votes": 1, "replies": [] }, { "id": 2330363, "author_name": "", "author_url": "", "post_date": "2023-07-04T23:40:42.900000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "2288554": "# Unet Baseline using PyTorch - [LB - 0.628]\n\nI've created a Unet baseline for this competition. It achieves a score of 0.580 on the leaderboard.\n\n**Training Notebook**: [Simple Unet Baseline (Train)](https://www.kaggle.com/code/shashwatraman/simple-unet-pytorch-baseline-train)\n**Inference Notebook**: [Simple Unet Baseline (Infer)](https://www.kaggle.com/code/shashwatraman/simple-unet-pytorch-baseline-infer)\n**Dataset Notebook** : [Contrails Dataset (Ash Color)](https://www.kaggle.com/code/shashwatraman/contrails-dataset-ash-color)\n\nTo get a better understanding of how everything works, please do look at the notebook used to create the dataset.\n\nSome points:\n- Library: Smp\n- Data: Ash Color images (With only the labeled frames and human_pixel_masks)\n- Backbone: EfficientNet-B0\n- Postprocessing: Finding the best threshold\n\nPlease upvote if you find this useful. If you don't understand anything, please do ask me, I'm really happy to help :)\n\nUpdate\nUsing EfficientNet B3, adding some augmentations and training for 30 epochs, the new version of the baseline gets a lb score of 0.628.\n\n\n\n", "2336638": "Update:\nUsing EfficientNet B3, adding some augmentations and training for 30 epochs, the new version of the baseline gets a lb score of 0.628.", "2292142": "Very clean, nice !👍", "2330363": "" } }