![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F84564%2F8c5c8b54444d6cbcfd7fdc85210d4cc4%2FUntitled.png?generation=1563757674140866&alt=media\")
I have used a site 1 image and trained the pre-trained model. But I am getting low accuracy in LB. The dataset contains two different site images.
Please share any reference available. Thanks in advance🙏
", "messages": [ { "id": "581460", "postDate": "07/22/2019 01:10:10", "content": "I have used a site 1 image and trained the pre-trained model. But I am getting low accuracy in LB. The dataset contains two different site images.
Please share any reference available. Thanks in advance🙏
", "rawMarkdown": "I have used a site 1 image and trained the pre-trained model. But I am getting low accuracy in LB. The dataset contains two different site images. \n\n Please share any reference available. Thanks in advance🙏", "votes": null }, { "id": "582340", "postDate": "07/23/2019 04:43:30", "content": "You are supposed to use both sites for training the model and then predict the labels for the test images, you can join the respective prediction column with submission file on \"id_code\". You can see a example here in Alexander Khar's kernel
", "rawMarkdown": "You are supposed to use both sites for training the model and then predict the labels for the test images, you can join the respective prediction column with submission file on \"id_code\". You can see a example here in [Alexander Khar's kernel](https://www.kaggle.com/alexanderkhar/transfer-learning-keras-starter-by-alex-khar)", "votes": null }, { "id": "582387", "postDate": "07/23/2019 05:54:58", "content": "Thank you Gajendra
", "rawMarkdown": "Thank you Gajendra", "votes": null }, { "id": "582614", "postDate": "07/23/2019 11:20:11", "content": "@sarques But how to stack them? Should i merge them into a single 12-channeled image or somehow else?
", "rawMarkdown": "sarques But how to stack them? Should i merge them into a single 12-channeled image or somehow else?", "votes": null }, { "id": "582652", "postDate": "07/23/2019 12:05:45", "content": "No, you should train your model on two different 6-channel images for each \"id_code\" and then predict your label on the test data. You can refer to the kernel provided in other comment (Alexander Khar), you can use that as your baseline model too.\nAlso you have two options to train your model, either change your 6-channel image to RGB or train it on 6-channel image itself.\nHope that helps. :)
", "rawMarkdown": "No, you should train your model on two different 6-channel images for each \"id_code\" and then predict your label on the test data. You can refer to the kernel provided in other comment (Alexander Khar), you can use that as your baseline model too.\nAlso you have two options to train your model, either change your 6-channel image to RGB or train it on 6-channel image itself.\nHope that helps. :)", "votes": null } ], "comments": [ { "id": 582340, "author_name": "sarques", "author_url": "", "post_date": "07/23/2019 04:43:30", "content": "You are supposed to use both sites for training the model and then predict the labels for the test images, you can join the respective prediction column with submission file on \"id_code\". You can see a example here in Alexander Khar's kernel
", "votes": null, "replies": [] }, { "id": 582387, "author_name": "sabarinathan", "author_url": "", "post_date": "07/23/2019 05:54:58", "content": "Thank you Gajendra
", "votes": null, "replies": [] }, { "id": 582614, "author_name": "rafailfridman", "author_url": "", "post_date": "07/23/2019 11:20:11", "content": "@sarques But how to stack them? Should i merge them into a single 12-channeled image or somehow else?
", "votes": null, "replies": [ { "id": 582652, "author_name": "sarques", "author_url": "", "post_date": "07/23/2019 12:05:45", "content": "No, you should train your model on two different 6-channel images for each \"id_code\" and then predict your label on the test data. You can refer to the kernel provided in other comment (Alexander Khar), you can use that as your baseline model too.\nAlso you have two options to train your model, either change your 6-channel image to RGB or train it on 6-channel image itself.\nHope that helps. :)
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "581460": "I have used a site 1 image and trained the pre-trained model. But I am getting low accuracy in LB. The dataset contains two different site images. \n\n Please share any reference available. Thanks in advance🙏", "582340": "You are supposed to use both sites for training the model and then predict the labels for the test images, you can join the respective prediction column with submission file on \"id_code\". You can see a example here in [Alexander Khar's kernel](https://www.kaggle.com/alexanderkhar/transfer-learning-keras-starter-by-alex-khar)", "582387": "Thank you Gajendra", "582614": "sarques But how to stack them? Should i merge them into a single 12-channeled image or somehow else?", "582652": "No, you should train your model on two different 6-channel images for each \"id_code\" and then predict your label on the test data. You can refer to the kernel provided in other comment (Alexander Khar), you can use that as your baseline model too.\nAlso you have two options to train your model, either change your 6-channel image to RGB or train it on 6-channel image itself.\nHope that helps. :)" }, "source": "meta" }