{ "id": 307317, "title": "🖼️ Coloring black and white images in dataset ", "url": "/competitions/happy-whale-and-dolphin/discussion/307317", "author_name": "", "post_date": "2022-02-13T18:47:16.294479200Z", "votes": 6, "comment_count": 3, "views": 0, "content": "

While talking about how can we handle single channeled images with @init27 we discussed what if we colorized the images with DeOldify

\n

Results from deoldify

\n

\"\"

\n

Github link: https://github.com/jantic/DeOldify

\n

notebook for implementation: notebook link

\n

Another running version is already converting black and white to colorized images here
\ncurrently running notebook link

\n

Some more results

\n

\"\"

\n

Dataset link:

\n

https://www.kaggle.com/somesh88/deoldify-single-channeled-converted-images

\n
\n

the dataset is uploaded completely 😊

\n
", "messages": [ { "id": "1688667", "postDate": "02/13/2022 18:47:16", "content": "

While talking about how can we handle single channeled images with @init27 we discussed what if we colorized the images with DeOldify

\n

Results from deoldify

\n

\"\"

\n

Github link: https://github.com/jantic/DeOldify

\n

notebook for implementation: notebook link

\n

Another running version is already converting black and white to colorized images here
\ncurrently running notebook link

\n

Some more results

\n

\"\"

\n

Dataset link:

\n

https://www.kaggle.com/somesh88/deoldify-single-channeled-converted-images

\n
\n

the dataset is uploaded completely 😊

\n
", "rawMarkdown": "While talking about how can we handle single channeled images with @init27 we discussed what if we colorized the images with DeOldify \n\n# Results from deoldify\n![](https://i.ibb.co/bQGF5DG/image.png)\n\n\n\nGithub link: [https://github.com/jantic/DeOldify](https://github.com/jantic/DeOldify)\n\nnotebook for implementation: [notebook link](https://www.kaggle.com/somesh88/coloring-single-channel-images-deoldify?scriptVersionId=87747808)\n\nAnother running version is already converting black and white to colorized images here \n[currently running notebook link ](https://www.kaggle.com/somesh88/coloring-single-channel-images-deoldify?scriptVersionId=87747745)\n\n\n## Some more results \n![](https://i.ibb.co/kKT3H3N/image.png)\n\n## Dataset link:\n[https://www.kaggle.com/somesh88/deoldify-single-channeled-converted-images]( https://www.kaggle.com/somesh88/deoldify-single-channeled-converted-images)\n> the dataset is uploaded completely 😊", "votes": null }, { "id": "1688753", "postDate": "02/13/2022 19:57:51", "content": "

DeOldify as Data Augmentation! 👌

\n

I'm not sure what might be the best way to evaluate this, maybe this ends up performing better since this is more I/P for a model to learn from compared to a B/W image. 🤔

", "rawMarkdown": "DeOldify as Data Augmentation! 👌\n\nI'm not sure what might be the best way to evaluate this, maybe this ends up performing better since this is more I/P for a model to learn from compared to a B/W image. 🤔", "votes": null }, { "id": "1689196", "postDate": "02/14/2022 04:40:49", "content": "

Yes

\n

I cannot create dataset because of 9 hr limit :(
\nI tried but it failed :( ( 9 GPU hours gone just like that haha )

", "rawMarkdown": "Yes \n\nI cannot create dataset because of 9 hr limit :( \nI tried but it failed :( ( 9 GPU hours gone just like that haha )", "votes": null }, { "id": "1689197", "postDate": "02/14/2022 04:42:43", "content": "

Noo I was mistaken it ran it saved all b_and_w train images it threw error on test images

", "rawMarkdown": "Noo I was mistaken it ran it saved all b_and_w train images it threw error on test images", "votes": null } ], "comments": [ { "id": 1688753, "author_name": "init27", "author_url": "", "post_date": "02/13/2022 19:57:51", "content": "

DeOldify as Data Augmentation! 👌

\n

I'm not sure what might be the best way to evaluate this, maybe this ends up performing better since this is more I/P for a model to learn from compared to a B/W image. 🤔

", "votes": null, "replies": [ { "id": 1689196, "author_name": "somesh88", "author_url": "", "post_date": "02/14/2022 04:40:49", "content": "

Yes

\n

I cannot create dataset because of 9 hr limit :(
\nI tried but it failed :( ( 9 GPU hours gone just like that haha )

", "votes": null, "replies": [] }, { "id": 1689197, "author_name": "somesh88", "author_url": "", "post_date": "02/14/2022 04:42:43", "content": "

Noo I was mistaken it ran it saved all b_and_w train images it threw error on test images

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1688667": "While talking about how can we handle single channeled images with @init27 we discussed what if we colorized the images with DeOldify \n\n# Results from deoldify\n![](https://i.ibb.co/bQGF5DG/image.png)\n\n\n\nGithub link: [https://github.com/jantic/DeOldify](https://github.com/jantic/DeOldify)\n\nnotebook for implementation: [notebook link](https://www.kaggle.com/somesh88/coloring-single-channel-images-deoldify?scriptVersionId=87747808)\n\nAnother running version is already converting black and white to colorized images here \n[currently running notebook link ](https://www.kaggle.com/somesh88/coloring-single-channel-images-deoldify?scriptVersionId=87747745)\n\n\n## Some more results \n![](https://i.ibb.co/kKT3H3N/image.png)\n\n## Dataset link:\n[https://www.kaggle.com/somesh88/deoldify-single-channeled-converted-images]( https://www.kaggle.com/somesh88/deoldify-single-channeled-converted-images)\n> the dataset is uploaded completely 😊", "1688753": "DeOldify as Data Augmentation! 👌\n\nI'm not sure what might be the best way to evaluate this, maybe this ends up performing better since this is more I/P for a model to learn from compared to a B/W image. 🤔", "1689196": "Yes \n\nI cannot create dataset because of 9 hr limit :( \nI tried but it failed :( ( 9 GPU hours gone just like that haha )", "1689197": "Noo I was mistaken it ran it saved all b_and_w train images it threw error on test images" }, "source": "meta" }