Hello, I have some experience in deep learning, but feel that this is a great project to really sharpen my skills. I am currently stuck on how I should handle the large images for a CNN in pytorch. I setup a project that has a main function that will resize the images (currently using a couple of the thumbnails to speed up development) and then I use a simple CNN for training.
\nI would like to explore other ways to handle the large images using a CNN first, then later try other approaches. Can anyone help explain how I can do this?
", "messages": [ { "id": 2558189, "postDate": "2023-12-12T01:08:29.210Z", "content": "@ryanwaka Try to use a larger kernel in first layer with stride 2 (maybe you can try 7 for high resolution), you can try to use stride 2 to reduce the feature map width and height (In practice this better then pooling 2d), and you can try kernel 1 to reduce or expand the channel.
", "rawMarkdown": "@ryanwaka Try to use a larger kernel in first layer with stride 2 (maybe you can try 7 for high resolution), you can try to use stride 2 to reduce the feature map width and height (In practice this better then pooling 2d), and you can try kernel 1 to reduce or expand the channel. ", "votes": 1 }, { "id": 2557987, "postDate": "2023-12-11T19:31:25.097Z", "content": "Hello, I have some experience in deep learning, but feel that this is a great project to really sharpen my skills. I am currently stuck on how I should handle the large images for a CNN in pytorch. I setup a project that has a main function that will resize the images (currently using a couple of the thumbnails to speed up development) and then I use a simple CNN for training.
\nI would like to explore other ways to handle the large images using a CNN first, then later try other approaches. Can anyone help explain how I can do this?
", "rawMarkdown": "Hello, I have some experience in deep learning, but feel that this is a great project to really sharpen my skills. I am currently stuck on how I should handle the large images for a CNN in pytorch. I setup a project that has a main function that will resize the images (currently using a couple of the thumbnails to speed up development) and then I use a simple CNN for training. \n\nI would like to explore other ways to handle the large images using a CNN first, then later try other approaches. Can anyone help explain how I can do this?", "votes": 2 } ], "comments": [ { "id": 2558189, "author_name": "Ren-Di Wu", "author_url": "", "post_date": "2023-12-12T01:08:29.210000", "content": "@ryanwaka Try to use a larger kernel in first layer with stride 2 (maybe you can try 7 for high resolution), you can try to use stride 2 to reduce the feature map width and height (In practice this better then pooling 2d), and you can try kernel 1 to reduce or expand the channel.
", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "2558189": "@ryanwaka Try to use a larger kernel in first layer with stride 2 (maybe you can try 7 for high resolution), you can try to use stride 2 to reduce the feature map width and height (In practice this better then pooling 2d), and you can try kernel 1 to reduce or expand the channel. ", "2557987": "Hello, I have some experience in deep learning, but feel that this is a great project to really sharpen my skills. I am currently stuck on how I should handle the large images for a CNN in pytorch. I setup a project that has a main function that will resize the images (currently using a couple of the thumbnails to speed up development) and then I use a simple CNN for training. \n\nI would like to explore other ways to handle the large images using a CNN first, then later try other approaches. Can anyone help explain how I can do this?" } }