![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1528571%2Fe2795b974dfdb3a4edde2b53b0595c6e%2Fpipeline.png?generation=1590238645820972&alt=media\")
Inspired from the Alaska-I winners, I made a rough sketch of pipeline that can may help in this competition.
\n\n
Please let me know what you think about this
", "messages": [ { "id": "858402", "postDate": "05/23/2020 13:00:07", "content": "Inspired from the Alaska-I winners, I made a rough sketch of pipeline that can may help in this competition.
\n\n
Please let me know what you think about this
", "rawMarkdown": "Inspired from the [Alaska-I winners](http://www.ws.binghamton.edu/fridrich/Research/ALASKA-preprint1.pdf), I made a rough sketch of pipeline that can may help in this competition.\n\n\n\nPlease let me know what you think about this", "votes": null }, { "id": "858493", "postDate": "05/23/2020 14:29:55", "content": "may i missing smth but why you need both YCbCr and separated Y/Cb/Cr inputs?
", "rawMarkdown": "may i missing smth but why you need both YCbCr and separated Y/Cb/Cr inputs?", "votes": null }, { "id": "858537", "postDate": "05/23/2020 15:08:05", "content": "> why you need both YCbCr and separated Y/Cb/Cr inputs?
\n\nMaybe you don't. But there might be a possibility that YCbCr and separated Y/Cb/Cr might force the models to learn different representations, which cannot be learned otherwise.
\n\n>We hypothesize that when supplying all three channels, it is possible that the SRNet focuses on leveraging embedding inconsistencies between the luminance and the two chrominances, ignoring\npossibly useful but perhaps comparatively weaker signals that exist within each channel that are left “untapped” when training the YCrCb-SRNet. Training the SRNet only on one chrominance may\nforce the network to leverage different types of embedding artifacts.
", "rawMarkdown": "> why you need both YCbCr and separated Y/Cb/Cr inputs?\n\nMaybe you don't. But there might be a possibility that YCbCr and separated Y/Cb/Cr might force the models to learn different representations, which cannot be learned otherwise. \n\n>We hypothesize that when supplying all three channels, it is possible that the SRNet focuses on leveraging embedding inconsistencies between the luminance and the two chrominances, ignoring\npossibly useful but perhaps comparatively weaker signals that exist within each channel that are left “untapped” when training the YCrCb-SRNet. Training the SRNet only on one chrominance may\nforce the network to leverage different types of embedding artifacts.", "votes": null }, { "id": "883335", "postDate": "06/12/2020 14:43:01", "content": "ResNet/EffNets which was useful for which one of the format can you explain?
", "rawMarkdown": "ResNet/EffNets which was useful for which one of the format can you explain?", "votes": null } ], "comments": [ { "id": 858493, "author_name": "moewie94", "author_url": "", "post_date": "05/23/2020 14:29:55", "content": "may i missing smth but why you need both YCbCr and separated Y/Cb/Cr inputs?
", "votes": null, "replies": [ { "id": 858537, "author_name": "bibek777", "author_url": "", "post_date": "05/23/2020 15:08:05", "content": "> why you need both YCbCr and separated Y/Cb/Cr inputs?
\n\nMaybe you don't. But there might be a possibility that YCbCr and separated Y/Cb/Cr might force the models to learn different representations, which cannot be learned otherwise.
\n\n>We hypothesize that when supplying all three channels, it is possible that the SRNet focuses on leveraging embedding inconsistencies between the luminance and the two chrominances, ignoring\npossibly useful but perhaps comparatively weaker signals that exist within each channel that are left “untapped” when training the YCrCb-SRNet. Training the SRNet only on one chrominance may\nforce the network to leverage different types of embedding artifacts.
", "votes": null, "replies": [] } ] }, { "id": 883335, "author_name": "tasnimnishatislam", "author_url": "", "post_date": "06/12/2020 14:43:01", "content": "ResNet/EffNets which was useful for which one of the format can you explain?
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "858402": "Inspired from the [Alaska-I winners](http://www.ws.binghamton.edu/fridrich/Research/ALASKA-preprint1.pdf), I made a rough sketch of pipeline that can may help in this competition.\n\n\n\nPlease let me know what you think about this", "858493": "may i missing smth but why you need both YCbCr and separated Y/Cb/Cr inputs?", "858537": "> why you need both YCbCr and separated Y/Cb/Cr inputs?\n\nMaybe you don't. But there might be a possibility that YCbCr and separated Y/Cb/Cr might force the models to learn different representations, which cannot be learned otherwise. \n\n>We hypothesize that when supplying all three channels, it is possible that the SRNet focuses on leveraging embedding inconsistencies between the luminance and the two chrominances, ignoring\npossibly useful but perhaps comparatively weaker signals that exist within each channel that are left “untapped” when training the YCrCb-SRNet. Training the SRNet only on one chrominance may\nforce the network to leverage different types of embedding artifacts.", "883335": "ResNet/EffNets which was useful for which one of the format can you explain?" }, "source": "meta" }