We are back with the fifth edition of Kaggle BIPOC Mentor Stories . This is series which captures the Kaggling journey of myself and my mentee Muskan Jain .
\nHear it from Muskan on her learning experience in this competition
\nPlant Pathology 2021- FGVC8 was my first Kaggle competition related to image detection and also the first time I worked on this kind of problem. Here are the few things I learned from the competition.
\n1)EDA is an important part of this competition, particularly knowledge about the distribution of various colors (RGB)
\n2) Since it’s an image competition, another important part is image processing which includes canny image detection
\n3) To increase the dataset we can use data augmentation techniques like vertically and horizontally flipping the images, convolution(introduces a sunlight effect), blurring, etc.
\n4) Since this is an image competition, one would start with convolutional neural networks whose common building blocks are a convolutional layer(Conv2D), a max-pooling layer(MaxPool), and a nonlinear activation function(Eg:-ReLu). Also one would need to set up a TPU(Tensor Processing Unit) for this competition.
\n5) One can use a lot of different models like DenseNet, EfficientNet which can give a quite high accuracy and even ensemble, stack models, or use other strong validation techniques to generate better predictions. Moreover, since it is easy to find from the images if a leaf is healthy or unhealthy, one can cross-check their predictions(at least a few).
\nThanks and Regards,
\nTensor GIrl and Muskan Jain
", "messages": [ { "id": "1333755", "postDate": "06/03/2021 03:23:05", "content": "We are back with the fifth edition of Kaggle BIPOC Mentor Stories . This is series which captures the Kaggling journey of myself and my mentee Muskan Jain .
\nHear it from Muskan on her learning experience in this competition
\nPlant Pathology 2021- FGVC8 was my first Kaggle competition related to image detection and also the first time I worked on this kind of problem. Here are the few things I learned from the competition.
\n1)EDA is an important part of this competition, particularly knowledge about the distribution of various colors (RGB)
\n2) Since it’s an image competition, another important part is image processing which includes canny image detection
\n3) To increase the dataset we can use data augmentation techniques like vertically and horizontally flipping the images, convolution(introduces a sunlight effect), blurring, etc.
\n4) Since this is an image competition, one would start with convolutional neural networks whose common building blocks are a convolutional layer(Conv2D), a max-pooling layer(MaxPool), and a nonlinear activation function(Eg:-ReLu). Also one would need to set up a TPU(Tensor Processing Unit) for this competition.
\n5) One can use a lot of different models like DenseNet, EfficientNet which can give a quite high accuracy and even ensemble, stack models, or use other strong validation techniques to generate better predictions. Moreover, since it is easy to find from the images if a leaf is healthy or unhealthy, one can cross-check their predictions(at least a few).
\nThanks and Regards,
\nTensor GIrl and Muskan Jain
", "rawMarkdown": "We are back with the fifth edition of Kaggle BIPOC Mentor Stories . This is series which captures the Kaggling journey of myself and my mentee Muskan Jain .\n\nHear it from Muskan on her learning experience in this competition\n\n\n\n**Plant Pathology 2021- FGVC8 was my first Kaggle competition related to image detection and also the first time I worked on this kind of problem. Here are the few things I learned from the competition.**\n\n**1)EDA is an important part of this competition, particularly knowledge about the distribution of various colors (RGB)**\n\n**2) Since it’s an image competition, another important part is image processing which includes canny image detection**\n\n**3) To increase the dataset we can use data augmentation techniques like vertically and horizontally flipping the images, convolution(introduces a sunlight effect), blurring, etc.**\n\n**4) Since this is an image competition, one would start with convolutional neural networks whose common building blocks are a convolutional layer(Conv2D), a max-pooling layer(MaxPool), and a nonlinear activation function(Eg:-ReLu). Also one would need to set up a TPU(Tensor Processing Unit) for this competition.**\n\n**5) One can use a lot of different models like DenseNet, EfficientNet which can give a quite high accuracy and even ensemble, stack models, or use other strong validation techniques to generate better predictions. Moreover, since it is easy to find from the images if a leaf is healthy or unhealthy, one can cross-check their predictions(at least a few).**\n\nThanks and Regards,\n\nTensor GIrl and Muskan Jain", "votes": null }, { "id": "1367864", "postDate": "06/28/2021 06:24:58", "content": "Thanks for sharing, this is very helpful!
", "rawMarkdown": "Thanks for sharing, this is very helpful!", "votes": null } ], "comments": [ { "id": 1367864, "author_name": "saurabhbagchi", "author_url": "", "post_date": "06/28/2021 06:24:58", "content": "Thanks for sharing, this is very helpful!
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1333755": "We are back with the fifth edition of Kaggle BIPOC Mentor Stories . This is series which captures the Kaggling journey of myself and my mentee Muskan Jain .\n\nHear it from Muskan on her learning experience in this competition\n\n\n\n**Plant Pathology 2021- FGVC8 was my first Kaggle competition related to image detection and also the first time I worked on this kind of problem. Here are the few things I learned from the competition.**\n\n**1)EDA is an important part of this competition, particularly knowledge about the distribution of various colors (RGB)**\n\n**2) Since it’s an image competition, another important part is image processing which includes canny image detection**\n\n**3) To increase the dataset we can use data augmentation techniques like vertically and horizontally flipping the images, convolution(introduces a sunlight effect), blurring, etc.**\n\n**4) Since this is an image competition, one would start with convolutional neural networks whose common building blocks are a convolutional layer(Conv2D), a max-pooling layer(MaxPool), and a nonlinear activation function(Eg:-ReLu). Also one would need to set up a TPU(Tensor Processing Unit) for this competition.**\n\n**5) One can use a lot of different models like DenseNet, EfficientNet which can give a quite high accuracy and even ensemble, stack models, or use other strong validation techniques to generate better predictions. Moreover, since it is easy to find from the images if a leaf is healthy or unhealthy, one can cross-check their predictions(at least a few).**\n\nThanks and Regards,\n\nTensor GIrl and Muskan Jain", "1367864": "Thanks for sharing, this is very helpful!" }, "source": "meta" }