I'm new in deep learning field, my research is related to the topic of this competition \"Diabetic Retinopathy \". \n1- How can I improve myself in this field ?\n2- What are the main steps to work on this topic ?\n3- How can I achieve a progress in short time ?\n4- From where I can start coding to perform the needed algorithm ?\nThanks in advance. Good Luck for all of u :)
", "messages": [ { "id": "614718", "postDate": "09/01/2019 00:40:34", "content": "I'm new in deep learning field, my research is related to the topic of this competition \"Diabetic Retinopathy \". \n1- How can I improve myself in this field ?\n2- What are the main steps to work on this topic ?\n3- How can I achieve a progress in short time ?\n4- From where I can start coding to perform the needed algorithm ?\nThanks in advance. Good Luck for all of u :)
", "rawMarkdown": "I'm new in deep learning field, my research is related to the topic of this competition \"Diabetic Retinopathy \". \n1- How can I improve myself in this field ?\n2- What are the main steps to work on this topic ?\n3- How can I achieve a progress in short time ?\n4- From where I can start coding to perform the needed algorithm ?\nThanks in advance. Good Luck for all of u :)", "votes": null }, { "id": "614779", "postDate": "09/01/2019 03:59:43", "content": "Hi,
\n\nYou mentioned that you are new in deep learning, but your research is related to \"Diabetic Retinopathy\", then can I assume that you are from bio or medical background? If so, then you are already in advantage in terms of domain knowledge, compared with most of the participants. Now then only thing you need to pick up might be the machine learning/deep learning techniques. There are plenty of online sources available out there. You can start with forking some public kernels as a quick start, figuring out what people are doing in their codes.
\n\nNow back to your question.
\n\n1.-How can I improve myself in this field ?\nPick up some computer vision/image classification techniques. After that, apply your domain knowledge in \"Diabetic Retinopathy\" to improve those models. TBH, since I know nothing about \"Diabetic Retinopathy\", I just blindly apply those computer vision techniques to train my models. 😆
\n\n2- What are the main steps to work on this topic ?\nAt least for me, the steps are \n(a). load the images as numbers\n(b). pre-process images, resize to the same size + some cropping to cut out the black parts\n(c). train a deep learning network to classify the images into different levels of DR.\nOf course, with domain knowledge, you can have better pre-processing techniques to deal with the raw images.
\n\n3- How can I achieve a progress in short time ?\nFork some public kernels as a starter, modify their codes based on your understanding of the subject.
\n\n4- From where I can start coding to perform the needed algorithm ?\nSimilar to the above question, you don't have to start from scratch. Start from understanding other people's work, then add in what you want.
\n\nHope that helps.
", "rawMarkdown": "Hi,\n\nYou mentioned that you are new in deep learning, but your research is related to \"Diabetic Retinopathy\", then can I assume that you are from bio or medical background? If so, then you are already in advantage in terms of domain knowledge, compared with most of the participants. Now then only thing you need to pick up might be the machine learning/deep learning techniques. There are plenty of online sources available out there. You can start with forking some public kernels as a quick start, figuring out what people are doing in their codes.\n\nNow back to your question.\n\n1.-How can I improve myself in this field ?\nPick up some computer vision/image classification techniques. After that, apply your domain knowledge in \"Diabetic Retinopathy\" to improve those models. TBH, since I know nothing about \"Diabetic Retinopathy\", I just blindly apply those computer vision techniques to train my models. 😆 \n\n2- What are the main steps to work on this topic ?\nAt least for me, the steps are \n(a). load the images as numbers\n(b). pre-process images, resize to the same size + some cropping to cut out the black parts\n(c). train a deep learning network to classify the images into different levels of DR.\nOf course, with domain knowledge, you can have better pre-processing techniques to deal with the raw images.\n\n3- How can I achieve a progress in short time ?\nFork some public kernels as a starter, modify their codes based on your understanding of the subject.\n\n4- From where I can start coding to perform the needed algorithm ?\nSimilar to the above question, you don't have to start from scratch. Start from understanding other people's work, then add in what you want.\n\nHope that helps.", "votes": null }, { "id": "614821", "postDate": "09/01/2019 05:51:12", "content": "@liusiyuan already mentioned good tips, to add:\n- Read up on winning solutions to previous/related competitions, notably the 2015 competition that held the same topic essentially\n- Read up on papers regarding using DL to detect DR, which some of them are already mentioned in the discussion forum\n- Explore! It's important that you try out either the research papers' experiment and/or within this competition.
\n\nLastly, Kaggle competitions accept late submission after the competition ends. So, if you really want to learn, you can spend some good time to follow the winner's solutions when they're out (which a lot of them in the gold region and above post once it's over) to reproduce same or similar results as them. Best of luck.
", "rawMarkdown": "liusiyuan already mentioned good tips, to add:\n- Read up on winning solutions to previous/related competitions, notably the 2015 competition that held the same topic essentially\n- Read up on papers regarding using DL to detect DR, which some of them are already mentioned in the discussion forum\n- Explore! It's important that you try out either the research papers' experiment and/or within this competition.\n\nLastly, Kaggle competitions accept late submission after the competition ends. So, if you really want to learn, you can spend some good time to follow the winner's solutions when they're out (which a lot of them in the gold region and above post once it's over) to reproduce same or similar results as them. Best of luck.", "votes": null }, { "id": "615524", "postDate": "09/02/2019 03:51:08", "content": "Many thanks. You really helped me a lot :)
", "rawMarkdown": "Many thanks. You really helped me a lot :)", "votes": null }, { "id": "615526", "postDate": "09/02/2019 03:53:47", "content": "Thanks, You are really helpful :)
", "rawMarkdown": "Thanks, You are really helpful :)", "votes": null } ], "comments": [ { "id": 614779, "author_name": "liusiyuan", "author_url": "", "post_date": "09/01/2019 03:59:43", "content": "Hi,
\n\nYou mentioned that you are new in deep learning, but your research is related to \"Diabetic Retinopathy\", then can I assume that you are from bio or medical background? If so, then you are already in advantage in terms of domain knowledge, compared with most of the participants. Now then only thing you need to pick up might be the machine learning/deep learning techniques. There are plenty of online sources available out there. You can start with forking some public kernels as a quick start, figuring out what people are doing in their codes.
\n\nNow back to your question.
\n\n1.-How can I improve myself in this field ?\nPick up some computer vision/image classification techniques. After that, apply your domain knowledge in \"Diabetic Retinopathy\" to improve those models. TBH, since I know nothing about \"Diabetic Retinopathy\", I just blindly apply those computer vision techniques to train my models. 😆
\n\n2- What are the main steps to work on this topic ?\nAt least for me, the steps are \n(a). load the images as numbers\n(b). pre-process images, resize to the same size + some cropping to cut out the black parts\n(c). train a deep learning network to classify the images into different levels of DR.\nOf course, with domain knowledge, you can have better pre-processing techniques to deal with the raw images.
\n\n3- How can I achieve a progress in short time ?\nFork some public kernels as a starter, modify their codes based on your understanding of the subject.
\n\n4- From where I can start coding to perform the needed algorithm ?\nSimilar to the above question, you don't have to start from scratch. Start from understanding other people's work, then add in what you want.
\n\nHope that helps.
", "votes": null, "replies": [ { "id": 615524, "author_name": "nadahossam", "author_url": "", "post_date": "09/02/2019 03:51:08", "content": "Many thanks. You really helped me a lot :)
", "votes": null, "replies": [] } ] }, { "id": 614821, "author_name": "joonl04", "author_url": "", "post_date": "09/01/2019 05:51:12", "content": "@liusiyuan already mentioned good tips, to add:\n- Read up on winning solutions to previous/related competitions, notably the 2015 competition that held the same topic essentially\n- Read up on papers regarding using DL to detect DR, which some of them are already mentioned in the discussion forum\n- Explore! It's important that you try out either the research papers' experiment and/or within this competition.
\n\nLastly, Kaggle competitions accept late submission after the competition ends. So, if you really want to learn, you can spend some good time to follow the winner's solutions when they're out (which a lot of them in the gold region and above post once it's over) to reproduce same or similar results as them. Best of luck.
", "votes": null, "replies": [ { "id": 615526, "author_name": "nadahossam", "author_url": "", "post_date": "09/02/2019 03:53:47", "content": "Thanks, You are really helpful :)
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "614718": "I'm new in deep learning field, my research is related to the topic of this competition \"Diabetic Retinopathy \". \n1- How can I improve myself in this field ?\n2- What are the main steps to work on this topic ?\n3- How can I achieve a progress in short time ?\n4- From where I can start coding to perform the needed algorithm ?\nThanks in advance. Good Luck for all of u :)", "614779": "Hi,\n\nYou mentioned that you are new in deep learning, but your research is related to \"Diabetic Retinopathy\", then can I assume that you are from bio or medical background? If so, then you are already in advantage in terms of domain knowledge, compared with most of the participants. Now then only thing you need to pick up might be the machine learning/deep learning techniques. There are plenty of online sources available out there. You can start with forking some public kernels as a quick start, figuring out what people are doing in their codes.\n\nNow back to your question.\n\n1.-How can I improve myself in this field ?\nPick up some computer vision/image classification techniques. After that, apply your domain knowledge in \"Diabetic Retinopathy\" to improve those models. TBH, since I know nothing about \"Diabetic Retinopathy\", I just blindly apply those computer vision techniques to train my models. 😆 \n\n2- What are the main steps to work on this topic ?\nAt least for me, the steps are \n(a). load the images as numbers\n(b). pre-process images, resize to the same size + some cropping to cut out the black parts\n(c). train a deep learning network to classify the images into different levels of DR.\nOf course, with domain knowledge, you can have better pre-processing techniques to deal with the raw images.\n\n3- How can I achieve a progress in short time ?\nFork some public kernels as a starter, modify their codes based on your understanding of the subject.\n\n4- From where I can start coding to perform the needed algorithm ?\nSimilar to the above question, you don't have to start from scratch. Start from understanding other people's work, then add in what you want.\n\nHope that helps.", "614821": "liusiyuan already mentioned good tips, to add:\n- Read up on winning solutions to previous/related competitions, notably the 2015 competition that held the same topic essentially\n- Read up on papers regarding using DL to detect DR, which some of them are already mentioned in the discussion forum\n- Explore! It's important that you try out either the research papers' experiment and/or within this competition.\n\nLastly, Kaggle competitions accept late submission after the competition ends. So, if you really want to learn, you can spend some good time to follow the winner's solutions when they're out (which a lot of them in the gold region and above post once it's over) to reproduce same or similar results as them. Best of luck.", "615524": "Many thanks. You really helped me a lot :)", "615526": "Thanks, You are really helpful :)" }, "source": "meta" }