![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2034058%2F6553d71008d94c2c23886b7884bc3a00%2Fcutmix.PNG?generation=1582358733604782&alt=media\")
i tried to use cutmix/mixup using code shared in this POST
\n\ni tried mixup and it works fine for me,but when i tried mixup and cutmix all together i got this error : \n
without changing CNN architecture,where to change in cutmix to make this work for single channel images? it would be highly appreciated if you share cutmix/mixup pytorch implementation for single channel images,thanks in advance.
\n\ni tried to use cutmix/mixup using code shared in this POST
\n\ni tried mixup and it works fine for me,but when i tried mixup and cutmix all together i got this error : \n
without changing CNN architecture,where to change in cutmix to make this work for single channel images? it would be highly appreciated if you share cutmix/mixup pytorch implementation for single channel images,thanks in advance.
\n\nThe input dimensions should be BxCxHxW, even in the case of 1 channel: Bx1xHxW. Check your input dimension, I think the channel dim is missing (BxHxW). In this case try something like this: input = input.unsqueeze(1)\nOr if input is ndarray instead of tensor:input = np.expand_dims(input, axis=1)
Hi @pestipeti \nthank you for your reply,if you check the screenshot carefully you can see i have used this : \noutputs1,outputs2,outputs3 = model(inputs.unsqueeze(1).float())
\n\nbut for mixup i don't get any error,i get error only for cutmix
", "rawMarkdown": "Hi @pestipeti \nthank you for your reply,if you check the screenshot carefully you can see i have used this : \n**outputs1,outputs2,outputs3 = model(inputs.unsqueeze(1).float())**\n\nbut for mixup i don't get any error,i get error only for cutmix", "votes": null }, { "id": "753562", "postDate": "02/22/2020 11:17:38", "content": "cutmix needs the proper dimensions as well, you have to do unsqueeze before cutmix.
", "rawMarkdown": "cutmix needs the proper dimensions as well, you have to do unsqueeze before cutmix.", "votes": null }, { "id": "753581", "postDate": "02/22/2020 12:15:57", "content": "@pestipeti thank you\ni will give it a try and will let you know
", "rawMarkdown": "pestipeti thank you\ni will give it a try and will let you know", "votes": null }, { "id": "753776", "postDate": "02/22/2020 16:33:42", "content": "it works @pestipeti \nthank you a lot for your help
", "rawMarkdown": "it works @pestipeti \nthank you a lot for your help", "votes": null }, { "id": "753781", "postDate": "02/22/2020 16:41:06", "content": "You're welcome.
", "rawMarkdown": "You're welcome.", "votes": null } ], "comments": [ { "id": 753465, "author_name": "pestipeti", "author_url": "", "post_date": "02/22/2020 08:12:39", "content": "The input dimensions should be BxCxHxW, even in the case of 1 channel: Bx1xHxW. Check your input dimension, I think the channel dim is missing (BxHxW). In this case try something like this: input = input.unsqueeze(1)\nOr if input is ndarray instead of tensor:input = np.expand_dims(input, axis=1)
Hi @pestipeti \nthank you for your reply,if you check the screenshot carefully you can see i have used this : \noutputs1,outputs2,outputs3 = model(inputs.unsqueeze(1).float())
\n\nbut for mixup i don't get any error,i get error only for cutmix
", "votes": null, "replies": [] }, { "id": 753562, "author_name": "pestipeti", "author_url": "", "post_date": "02/22/2020 11:17:38", "content": "cutmix needs the proper dimensions as well, you have to do unsqueeze before cutmix.
", "votes": null, "replies": [] }, { "id": 753581, "author_name": "mobassir", "author_url": "", "post_date": "02/22/2020 12:15:57", "content": "@pestipeti thank you\ni will give it a try and will let you know
", "votes": null, "replies": [] }, { "id": 753776, "author_name": "mobassir", "author_url": "", "post_date": "02/22/2020 16:33:42", "content": "it works @pestipeti \nthank you a lot for your help
", "votes": null, "replies": [] }, { "id": 753781, "author_name": "pestipeti", "author_url": "", "post_date": "02/22/2020 16:41:06", "content": "You're welcome.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "753460": "i tried to use cutmix/mixup using code shared in this [POST](https://www.kaggle.com/c/bengaliai-cv19/discussion/126504) \n\ni tried mixup and it works fine for me,but when i tried mixup and cutmix all together i got this error : \n\n\nwithout changing CNN architecture,where to change in cutmix to make this work for single channel images? it would be highly appreciated if you share cutmix/mixup pytorch implementation for single channel images,thanks in advance.\n#Happy_kaggling :)", "753465": "The input dimensions should be BxCxHxW, even in the case of 1 channel: Bx1xHxW. Check your input dimension, I think the channel dim is missing (BxHxW). In this case try something like this: `input = input.unsqueeze(1)`\nOr if input is ndarray instead of tensor:`input = np.expand_dims(input, axis=1)`", "753559": "Hi @pestipeti \nthank you for your reply,if you check the screenshot carefully you can see i have used this : \n**outputs1,outputs2,outputs3 = model(inputs.unsqueeze(1).float())**\n\nbut for mixup i don't get any error,i get error only for cutmix", "753562": "cutmix needs the proper dimensions as well, you have to do unsqueeze before cutmix.", "753581": "pestipeti thank you\ni will give it a try and will let you know", "753776": "it works @pestipeti \nthank you a lot for your help", "753781": "You're welcome." }, "source": "meta" }