{ "id": 211201, "title": "Deit Model CV - 0.94 LB - 0.889", "url": "/competitions/cassava-leaf-disease-classification/discussion/211201", "author_name": "VR PIRATES", "post_date": "2021-01-14T07:48:27.283000", "votes": 2, "comment_count": 8, "views": 0, "content": "

Eff Net B3 Cv-0.897 lb - 0.897 but in diet model gap between cv and lb is too much. Can anyone help me with this?

", "messages": [ { "id": 1154409, "postDate": "2021-01-15T16:09:55.643Z", "content": "

One thing is for sure vision transformer models need lots of data, make sure you have much more augumentations.

", "rawMarkdown": "One thing is for sure vision transformer models need lots of data, make sure you have much more augumentations.", "votes": 1 }, { "id": 1152482, "postDate": "2021-01-14T07:48:27.283Z", "content": "

Eff Net B3 Cv-0.897 lb - 0.897 but in diet model gap between cv and lb is too much. Can anyone help me with this?

", "rawMarkdown": "Eff Net B3 Cv-0.897 lb - 0.897 but in diet model gap between cv and lb is too much. Can anyone help me with this?", "votes": 1 }, { "id": 1172990, "postDate": "2021-01-27T16:32:20.353Z", "content": "

CV = 0.94 means their is definitely leak somewhere.

\n

Particularly given the LB score.

\n

Check the data you're using on validation.

", "rawMarkdown": "CV = 0.94 means their is definitely leak somewhere. \n\nParticularly given the LB score.\n\n\nCheck the data you're using on validation." }, { "id": 1172969, "postDate": "2021-01-27T16:22:22.250Z", "content": "

@vrpirates did u face the error while loading pretrained weights from torch.hub.load
\nOnly one file(not dir) is allowed in the zipfile

", "rawMarkdown": "@vrpirates did u face the error while loading pretrained weights from torch.hub.load\n`Only one file(not dir) is allowed in the zipfile`" }, { "id": 1152649, "postDate": "2021-01-14T10:45:06.250Z", "content": "

I am surprised that Deit got a CV of 0.94. How did you train and validate? Resize or RandomResizedCrop? thanks~

", "rawMarkdown": "I am surprised that Deit got a CV of 0.94. How did you train and validate? Resize or RandomResizedCrop? thanks~", "replies": [ { "id": 1152728, "postDate": "2021-01-14T11:57:40.003Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 1152645, "postDate": "2021-01-14T10:40:56.367Z", "content": "

we cannot get pre-trained models with 512*512.
\nFor me, I haven't test my Deit model on Lb. I just trained it but showed a low CV score.

", "rawMarkdown": "we cannot get pre-trained models with 512*512. \nFor me, I haven't test my Deit model on Lb. I just trained it but showed a low CV score." }, { "id": 1152632, "postDate": "2021-01-14T10:32:18.007Z", "content": "

do you train Deit with a size 224*224?

", "rawMarkdown": "do you train Deit with a size 224*224?", "replies": [ { "id": 1152634, "postDate": "2021-01-14T10:34:48.373Z", "content": "

yes, Should i train with 512*512?

", "rawMarkdown": "yes, Should i train with 512*512?" } ] } ], "comments": [ { "id": 1154409, "author_name": "Ankush kuwar", "author_url": "", "post_date": "2021-01-15T16:09:55.643000", "content": "

One thing is for sure vision transformer models need lots of data, make sure you have much more augumentations.

", "votes": 1, "replies": [] }, { "id": 1172990, "author_name": "Serigne ", "author_url": "", "post_date": "2021-01-27T16:32:20.353000", "content": "

CV = 0.94 means their is definitely leak somewhere.

\n

Particularly given the LB score.

\n

Check the data you're using on validation.

", "votes": 0, "replies": [] }, { "id": 1172969, "author_name": "Jaideep", "author_url": "", "post_date": "2021-01-27T16:22:22.250000", "content": "

@vrpirates did u face the error while loading pretrained weights from torch.hub.load
\nOnly one file(not dir) is allowed in the zipfile

", "votes": 0, "replies": [] }, { "id": 1152649, "author_name": "Qixin Yan", "author_url": "", "post_date": "2021-01-14T10:45:06.250000", "content": "

I am surprised that Deit got a CV of 0.94. How did you train and validate? Resize or RandomResizedCrop? thanks~

", "votes": 0, "replies": [ { "id": 1152728, "author_name": "", "author_url": "", "post_date": "2021-01-14T11:57:40.003000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 1152645, "author_name": "Qixin Yan", "author_url": "", "post_date": "2021-01-14T10:40:56.367000", "content": "

we cannot get pre-trained models with 512*512.
\nFor me, I haven't test my Deit model on Lb. I just trained it but showed a low CV score.

", "votes": 0, "replies": [] }, { "id": 1152632, "author_name": "Qixin Yan", "author_url": "", "post_date": "2021-01-14T10:32:18.007000", "content": "

do you train Deit with a size 224*224?

", "votes": 0, "replies": [ { "id": 1152634, "author_name": "VR PIRATES", "author_url": "", "post_date": "2021-01-14T10:34:48.373000", "content": "

yes, Should i train with 512*512?

", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1154409": "One thing is for sure vision transformer models need lots of data, make sure you have much more augumentations.", "1152482": "Eff Net B3 Cv-0.897 lb - 0.897 but in diet model gap between cv and lb is too much. Can anyone help me with this?", "1172990": "CV = 0.94 means their is definitely leak somewhere. \n\nParticularly given the LB score.\n\n\nCheck the data you're using on validation.", "1172969": "@vrpirates did u face the error while loading pretrained weights from torch.hub.load\n`Only one file(not dir) is allowed in the zipfile`", "1152649": "I am surprised that Deit got a CV of 0.94. How did you train and validate? Resize or RandomResizedCrop? thanks~", "1152645": "we cannot get pre-trained models with 512*512. \nFor me, I haven't test my Deit model on Lb. I just trained it but showed a low CV score.", "1152632": "do you train Deit with a size 224*224?" } }