I am thinking about using Semantic semgentation and classifying each pixel into one of the categories. It it a good idea??
", "messages": [ { "id": "1298961", "postDate": "05/09/2021 11:13:13", "content": "I am thinking about using Semantic semgentation and classifying each pixel into one of the categories. It it a good idea??
", "rawMarkdown": "I am thinking about using Semantic semgentation and classifying each pixel into one of the categories. It it a good idea??", "votes": null }, { "id": "1299203", "postDate": "05/09/2021 14:35:33", "content": "How can you train the model without segmentation mask?
", "rawMarkdown": "How can you train the model without segmentation mask?", "votes": null }, { "id": "1299303", "postDate": "05/09/2021 15:28:35", "content": "i have already sarted creating masks using online image annotator. But point is will it be useful or not.
", "rawMarkdown": "i have already sarted creating masks using online image annotator. But point is will it be useful or not.", "votes": null }, { "id": "1299394", "postDate": "05/09/2021 16:17:08", "content": "You mean you can identify the diseases for each pixel?
\nI think it's very difficult.
I think the approach using mask will be really difficult, one thing you can obviously try is ,using the mask , segment the leaf only which has a disease(or healthy) then, using the co-ordinates of the mask you can extract the leaf only in RGB format and finally use any model on that extracted image. It will probably help in training the model faster and might give more accurate results. Please do let me know if you tried this method and if it worked or not.
\nThank you
Yes, but the results will be better than classification. Actually I am working on a research project for this, last year for PLant PAth 2020 also i tried, the results were better than even ensembling the models.
", "rawMarkdown": "Yes, but the results will be better than classification. Actually I am working on a research project for this, last year for PLant PAth 2020 also i tried, the results were better than even ensembling the models.", "votes": null }, { "id": "1303766", "postDate": "05/12/2021 08:24:23", "content": "yeah, those masks are improve the results but not much. Already tried this. Plant Village dataset has such masks.
", "rawMarkdown": "yeah, those masks are improve the results but not much. Already tried this. Plant Village dataset has such masks.", "votes": null }, { "id": "1303816", "postDate": "05/12/2021 09:04:02", "content": "Maybe you should train a segmentation model?
\nThis model is used to create a mask for only one main leaf by vision attention or all leaves?
\nClassification model is indeed subject to background interference.
\nI think you can try this.
\nThis should be easier than identify for each pixel.
I tried some Image Processing techniques and some of which actually showed some positive results . This is the notebook link for your reference, tell me if you find it useful.
\nKaggle Notebook:https://www.kaggle.com/amartyabhattacharya/beginner-s-image-processing-and-eda
I am working on a research project and my aim is to use segmentation. I created some masks myself and on unet getting good results.
\ncurious to know, Do you work on research projects?
Its good work, but i am looking for segmentation based approaches, my approach is working fine but now have to find some patterns and improve the results. I am talking about pipeline.
", "rawMarkdown": "Its good work, but i am looking for segmentation based approaches, my approach is working fine but now have to find some patterns and improve the results. I am talking about pipeline.", "votes": null } ], "comments": [ { "id": 1299203, "author_name": "rajatsanjaypatel", "author_url": "", "post_date": "05/09/2021 14:35:33", "content": "How can you train the model without segmentation mask?
", "votes": null, "replies": [ { "id": 1299303, "author_name": "shubhamdivakar", "author_url": "", "post_date": "05/09/2021 15:28:35", "content": "i have already sarted creating masks using online image annotator. But point is will it be useful or not.
", "votes": null, "replies": [] }, { "id": 1299394, "author_name": "rainyq", "author_url": "", "post_date": "05/09/2021 16:17:08", "content": "You mean you can identify the diseases for each pixel?
\nI think it's very difficult.
Yes, but the results will be better than classification. Actually I am working on a research project for this, last year for PLant PAth 2020 also i tried, the results were better than even ensembling the models.
", "votes": null, "replies": [] }, { "id": 1303816, "author_name": "rainyq", "author_url": "", "post_date": "05/12/2021 09:04:02", "content": "Maybe you should train a segmentation model?
\nThis model is used to create a mask for only one main leaf by vision attention or all leaves?
\nClassification model is indeed subject to background interference.
\nI think you can try this.
\nThis should be easier than identify for each pixel.
I am working on a research project and my aim is to use segmentation. I created some masks myself and on unet getting good results.
\ncurious to know, Do you work on research projects?
I think the approach using mask will be really difficult, one thing you can obviously try is ,using the mask , segment the leaf only which has a disease(or healthy) then, using the co-ordinates of the mask you can extract the leaf only in RGB format and finally use any model on that extracted image. It will probably help in training the model faster and might give more accurate results. Please do let me know if you tried this method and if it worked or not.
\nThank you
yeah, those masks are improve the results but not much. Already tried this. Plant Village dataset has such masks.
", "votes": null, "replies": [] } ] }, { "id": 1303839, "author_name": "amartyabhattacharya", "author_url": "", "post_date": "05/12/2021 09:19:49", "content": "I tried some Image Processing techniques and some of which actually showed some positive results . This is the notebook link for your reference, tell me if you find it useful.
\nKaggle Notebook:https://www.kaggle.com/amartyabhattacharya/beginner-s-image-processing-and-eda
Its good work, but i am looking for segmentation based approaches, my approach is working fine but now have to find some patterns and improve the results. I am talking about pipeline.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1298961": "I am thinking about using Semantic semgentation and classifying each pixel into one of the categories. It it a good idea??", "1299203": "How can you train the model without segmentation mask?", "1299303": "i have already sarted creating masks using online image annotator. But point is will it be useful or not.", "1299394": "You mean you can identify the diseases for each pixel?\nI think it's very difficult.", "1303433": "I think the approach using mask will be really difficult, one thing you can obviously try is ,using the mask , segment the leaf only which has a disease(or healthy) then, using the co-ordinates of the mask you can extract the leaf only in RGB format and finally use any model on that extracted image. It will probably help in training the model faster and might give more accurate results. Please do let me know if you tried this method and if it worked or not. \nThank you", "1303763": "Yes, but the results will be better than classification. Actually I am working on a research project for this, last year for PLant PAth 2020 also i tried, the results were better than even ensembling the models.", "1303766": "yeah, those masks are improve the results but not much. Already tried this. Plant Village dataset has such masks.", "1303816": "Maybe you should train a segmentation model?\nThis model is used to create a mask for only one main leaf by vision attention or all leaves?\nClassification model is indeed subject to background interference.\nI think you can try this.\nThis should be easier than identify for each pixel.", "1303839": "I tried some Image Processing techniques and some of which actually showed some positive results . This is the notebook link for your reference, tell me if you find it useful.\nKaggle Notebook:https://www.kaggle.com/amartyabhattacharya/beginner-s-image-processing-and-eda", "1305799": "I am working on a research project and my aim is to use segmentation. I created some masks myself and on unet getting good results. \ncurious to know, Do you work on research projects?", "1305804": "Its good work, but i am looking for segmentation based approaches, my approach is working fine but now have to find some patterns and improve the results. I am talking about pipeline." }, "source": "meta" }