![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F113660%2F6352999de5fba23ea0c091e2d08cb491%2FSelection_153.png?generation=1607418064732757&alt=media\")
Hello, everyone, i observe that there are many baselines treat one of the image as val dataset, doesn't that make validation unitary？ i try to crop all of the images and split them into train/ val dataset manually, but it seems doesn't get satisfactory result, can anyone offer some suggestions?
", "messages": [ { "id": 1105555, "postDate": "2020-12-08T02:02:39.710Z", "content": "Hello, everyone, i observe that there are many baselines treat one of the image as val dataset, doesn't that make validation unitary？ i try to crop all of the images and split them into train/ val dataset manually, but it seems doesn't get satisfactory result, can anyone offer some suggestions?
", "rawMarkdown": "Hello, everyone, i observe that there are many baselines treat one of the image as val dataset, doesn't that make validation unitary？ i try to crop all of the images and split them into train/ val dataset manually, but it seems doesn't get satisfactory result, can anyone offer some suggestions?", "votes": 2 }, { "id": 1105576, "postDate": "2020-12-08T02:51:47.410Z", "content": "I tried to use two images as the validation set, and the other as the training set. Compared with split after cropping, the cv of the former is more unstable
", "rawMarkdown": "I tried to use two images as the validation set, and the other as the training set. Compared with split after cropping, the cv of the former is more unstable", "replies": [ { "id": 1105578, "postDate": "2020-12-08T02:53:18.947Z", "content": "use one fresh frozen and one Formalin Fixed Paraffin Embedded (FFPE)
", "rawMarkdown": "use one fresh frozen and one Formalin Fixed Paraffin Embedded (FFPE)" }, { "id": 1105596, "postDate": "2020-12-08T03:06:29.417Z", "content": "Hi Heng, Thanks for your sharing, I don't know what is \"Formalin Fixed Paraffin Embedded \", would you mind share more about it
", "rawMarkdown": "Hi Heng, Thanks for your sharing, I don't know what is \"Formalin Fixed Paraffin Embedded \", would you mind share more about it" }, { "id": 1105628, "postDate": "2020-12-08T03:53:04.960Z", "content": "Thanks for your replay, so you use split after cropping at present? I have tried this split strategy, but the gap between CV and LB is about 8-9 points, i am confused about it.
", "rawMarkdown": "Thanks for your replay, so you use split after cropping at present? I have tried this split strategy, but the gap between CV and LB is about 8-9 points, i am confused about it.", "votes": 1 }, { "id": 1105643, "postDate": "2020-12-08T03:58:04.347Z", "content": "Thanks for your replay, could you please share code or more explanation about \"Formalin Fixed Paraffin Embedded \", i cannot find introduction about it on the internet?
", "rawMarkdown": "Thanks for your replay, could you please share code or more explanation about \"Formalin Fixed Paraffin Embedded \", i cannot find introduction about it on the internet?", "votes": 1 }, { "id": 1105759, "postDate": "2020-12-08T07:14:55.907Z", "content": "@yolo1996 I split training and validation by images, 0.75 for train, 0.25 for val
", "rawMarkdown": "@yolo1996 I split training and validation by images, 0.75 for train, 0.25 for val" }, { "id": 1105783, "postDate": "2020-12-08T07:33:48.563Z", "content": "thanks～，i will try again to split the train and val dataset
", "rawMarkdown": "thanks～，i will try again to split the train and val dataset" }, { "id": 1105854, "postDate": "2020-12-08T09:01:21.140Z", "content": "if you inspect the images closely, you can find two type of textures:
\n
I tried to use two images as the validation set, and the other as the training set. Compared with split after cropping, the cv of the former is more unstable
", "votes": 0, "replies": [ { "id": 1105578, "author_name": "hengck23", "author_url": "", "post_date": "2020-12-08T02:53:18.947000", "content": "use one fresh frozen and one Formalin Fixed Paraffin Embedded (FFPE)
", "votes": 0, "replies": [] }, { "id": 1105596, "author_name": "He", "author_url": "", "post_date": "2020-12-08T03:06:29.417000", "content": "Hi Heng, Thanks for your sharing, I don't know what is \"Formalin Fixed Paraffin Embedded \", would you mind share more about it
", "votes": 0, "replies": [] }, { "id": 1105628, "author_name": "JOJO", "author_url": "", "post_date": "2020-12-08T03:53:04.960000", "content": "Thanks for your replay, so you use split after cropping at present? I have tried this split strategy, but the gap between CV and LB is about 8-9 points, i am confused about it.
", "votes": 1, "replies": [] }, { "id": 1105643, "author_name": "JOJO", "author_url": "", "post_date": "2020-12-08T03:58:04.347000", "content": "Thanks for your replay, could you please share code or more explanation about \"Formalin Fixed Paraffin Embedded \", i cannot find introduction about it on the internet?
", "votes": 1, "replies": [] }, { "id": 1105759, "author_name": "He", "author_url": "", "post_date": "2020-12-08T07:14:55.907000", "content": "@yolo1996 I split training and validation by images, 0.75 for train, 0.25 for val
", "votes": 0, "replies": [] }, { "id": 1105783, "author_name": "JOJO", "author_url": "", "post_date": "2020-12-08T07:33:48.563000", "content": "thanks～，i will try again to split the train and val dataset
", "votes": 0, "replies": [] }, { "id": 1105854, "author_name": "hengck23", "author_url": "", "post_date": "2020-12-08T09:01:21.140000", "content": "if you inspect the images closely, you can find two type of textures:
\n