I found that LeafGAN was released only a few hours ago, https://github.com/IyatomiLab/LeafGAN!
\nThank you, Iyatomi lab!
I found that LeafGAN was released only a few hours ago, https://github.com/IyatomiLab/LeafGAN!
\nThank you, Iyatomi lab!
I am looking forward to seeing if it can make the competition funny. :)
", "rawMarkdown": "I am looking forward to seeing if it can make the competition funny. :)", "votes": null }, { "id": "1098751", "postDate": "12/01/2020 20:20:29", "content": "Great! Thanks for sharing. A minor concern:
\nWhen the images contain multiple and overlapping leaves, the LFLSeg fails to correctly segment the leaf area (last image of the full leaf cases). However, we do not expect the input which contains multiple leaves to be the case since we assume the input of the disease classifier is a single leaf image in this study.
But definitely their dataset can be used as an additional data!
", "rawMarkdown": "Great! Thanks for sharing. A minor concern: \n```When the images contain multiple and overlapping leaves, the LFLSeg fails to correctly segment the leaf area (last image of the full leaf cases). However, we do not expect the input which contains multiple leaves to be the case since we assume the input of the disease classifier is a single leaf image in this study.```\n\nBut definitely their dataset can be used as an additional data!", "votes": null }, { "id": "1099058", "postDate": "12/02/2020 03:52:55", "content": "This can be serious concern if anyone wants to use leafGAN for this competition as we have almost all images having multiple leaves.
", "rawMarkdown": "This can be serious concern if anyone wants to use leafGAN for this competition as we have almost all images having multiple leaves.", "votes": null }, { "id": "1099062", "postDate": "12/02/2020 04:02:11", "content": "Hey, I read the abstract from the paper and still don't get. How does it help the community?
", "rawMarkdown": "Hey, I read the abstract from the paper and still don't get. How does it help the community?", "votes": null }, { "id": "1099076", "postDate": "12/02/2020 04:25:53", "content": "Thank you for comment!
\nLeafGAN has been mentioned in
\nhttps://www.kaggle.com/kmat2019/cycle-gan-to-enlarge-training-data
\nand
\nhttps://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/199903.
Especially, @dimitreoliveira explained well in the discussion.
\nI also expect that LeafGAN can augment images to train models.
maybe good enhancement:
\nDifferentiable Augmentation for Data-Efficient GAN Training
\nhttps://arxiv.org/pdf/2006.10738.pdf
\" With DiffAugment, we achieve a state-of-the-art FID of 6.80 with an IS of 100.8 on ImageNet 128×128 and 2-4×
\nreductions of FID given 1,000 images on FFHQ and LSUN. Furthermore, with only 20% training data, we can match the top performance on CIFAR-10 and CIFAR-100.
Finally, our method can generate high-fidelity images using only 100 images without pre-training, while being on par with existing transfer learning algorithms.\"
", "rawMarkdown": "maybe good enhancement:\nDifferentiable Augmentation for Data-Efficient GAN Training\nhttps://arxiv.org/pdf/2006.10738.pdf\n\n\n\" With DiffAugment, we achieve a state-of-the-art FID of 6.80 with an IS of 100.8 on ImageNet 128×128 and 2-4×\nreductions of FID given 1,000 images on FFHQ and LSUN. Furthermore, with only 20% training data, we can match the top performance on CIFAR-10 and CIFAR-100.\n\nFinally, our method can generate high-fidelity images using only 100 images without pre-training, while being on par with existing transfer learning algorithms.\"", "votes": null } ], "comments": [ { "id": 1098554, "author_name": "piantic", "author_url": "", "post_date": "12/01/2020 17:42:20", "content": "I am looking forward to seeing if it can make the competition funny. :)
", "votes": null, "replies": [] }, { "id": 1098751, "author_name": "keremt", "author_url": "", "post_date": "12/01/2020 20:20:29", "content": "Great! Thanks for sharing. A minor concern:
\nWhen the images contain multiple and overlapping leaves, the LFLSeg fails to correctly segment the leaf area (last image of the full leaf cases). However, we do not expect the input which contains multiple leaves to be the case since we assume the input of the disease classifier is a single leaf image in this study.
But definitely their dataset can be used as an additional data!
", "votes": null, "replies": [ { "id": 1099058, "author_name": "kaushal2896", "author_url": "", "post_date": "12/02/2020 03:52:55", "content": "This can be serious concern if anyone wants to use leafGAN for this competition as we have almost all images having multiple leaves.
", "votes": null, "replies": [] } ] }, { "id": 1099062, "author_name": "vyombhatia", "author_url": "", "post_date": "12/02/2020 04:02:11", "content": "Hey, I read the abstract from the paper and still don't get. How does it help the community?
", "votes": null, "replies": [ { "id": 1099076, "author_name": "yosukeyama", "author_url": "", "post_date": "12/02/2020 04:25:53", "content": "Thank you for comment!
\nLeafGAN has been mentioned in
\nhttps://www.kaggle.com/kmat2019/cycle-gan-to-enlarge-training-data
\nand
\nhttps://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/199903.
Especially, @dimitreoliveira explained well in the discussion.
\nI also expect that LeafGAN can augment images to train models.
maybe good enhancement:
\nDifferentiable Augmentation for Data-Efficient GAN Training
\nhttps://arxiv.org/pdf/2006.10738.pdf
\" With DiffAugment, we achieve a state-of-the-art FID of 6.80 with an IS of 100.8 on ImageNet 128×128 and 2-4×
\nreductions of FID given 1,000 images on FFHQ and LSUN. Furthermore, with only 20% training data, we can match the top performance on CIFAR-10 and CIFAR-100.
Finally, our method can generate high-fidelity images using only 100 images without pre-training, while being on par with existing transfer learning algorithms.\"
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1098504": "I found that LeafGAN was released only a few hours ago, https://github.com/IyatomiLab/LeafGAN!\nThank you, Iyatomi lab!", "1098554": "I am looking forward to seeing if it can make the competition funny. :)", "1098751": "Great! Thanks for sharing. A minor concern: \n```When the images contain multiple and overlapping leaves, the LFLSeg fails to correctly segment the leaf area (last image of the full leaf cases). However, we do not expect the input which contains multiple leaves to be the case since we assume the input of the disease classifier is a single leaf image in this study.```\n\nBut definitely their dataset can be used as an additional data!", "1099058": "This can be serious concern if anyone wants to use leafGAN for this competition as we have almost all images having multiple leaves.", "1099062": "Hey, I read the abstract from the paper and still don't get. How does it help the community?", "1099076": "Thank you for comment!\n\nLeafGAN has been mentioned in \nhttps://www.kaggle.com/kmat2019/cycle-gan-to-enlarge-training-data\nand\nhttps://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/199903.\n\nEspecially, @dimitreoliveira explained well in the discussion.\nI also expect that LeafGAN can augment images to train models.", "1099285": "maybe good enhancement:\nDifferentiable Augmentation for Data-Efficient GAN Training\nhttps://arxiv.org/pdf/2006.10738.pdf\n\n\n\" With DiffAugment, we achieve a state-of-the-art FID of 6.80 with an IS of 100.8 on ImageNet 128×128 and 2-4×\nreductions of FID given 1,000 images on FFHQ and LSUN. Furthermore, with only 20% training data, we can match the top performance on CIFAR-10 and CIFAR-100.\n\nFinally, our method can generate high-fidelity images using only 100 images without pre-training, while being on par with existing transfer learning algorithms.\"" }, "source": "meta" }