{ "id": 246117, "title": "Dealing With Resizing Image boxes", "url": "/competitions/siim-covid19-detection/discussion/246117", "author_name": "", "post_date": "2021-06-14T02:54:13.409893500Z", "votes": 3, "comment_count": 2, "views": 0, "content": "

The Original Dimensions of the Images are in the size of around 4000x4000.This is too much and resizing is a must . My plan is to convert into PNG formats and exploring various resolutions .The problem here is that we also need to deal with the boxes coordinates but just dividing them by the resolution is not the correct way .The boxes i made by this method were too off the real ones .Any ideas on how to deal with this situation are welcome .
\nYou can view the difference in the image hosted in the link below.Resolution is 256x256 .
\n![https://ibb.co/b1jSyFN]

", "messages": [ { "id": "1348381", "postDate": "06/14/2021 02:54:13", "content": "

The Original Dimensions of the Images are in the size of around 4000x4000.This is too much and resizing is a must . My plan is to convert into PNG formats and exploring various resolutions .The problem here is that we also need to deal with the boxes coordinates but just dividing them by the resolution is not the correct way .The boxes i made by this method were too off the real ones .Any ideas on how to deal with this situation are welcome .
\nYou can view the difference in the image hosted in the link below.Resolution is 256x256 .
\n![https://ibb.co/b1jSyFN]

", "rawMarkdown": "The Original Dimensions of the Images are in the size of around 4000x4000.This is too much and resizing is a must . My plan is to convert into PNG formats and exploring various resolutions .The problem here is that we also need to deal with the boxes coordinates but just dividing them by the resolution is not the correct way .The boxes i made by this method were too off the real ones .Any ideas on how to deal with this situation are welcome .\nYou can view the difference in the image hosted in the link below.Resolution is 256x256 .\n![https://ibb.co/b1jSyFN]", "votes": null }, { "id": "1348654", "postDate": "06/14/2021 07:07:00", "content": "

You can use the Albumentations package to resize both the images and bounding boxes. It's surprisingly simple. This is a link to a quick tutorial:
\nhttps://sheldonsebastian94.medium.com/resizing-image-and-bounding-boxes-for-object-detection-7b9d9463125a

", "rawMarkdown": "You can use the Albumentations package to resize both the images and bounding boxes. It's surprisingly simple. This is a link to a quick tutorial:\nhttps://sheldonsebastian94.medium.com/resizing-image-and-bounding-boxes-for-object-detection-7b9d9463125a", "votes": null }, { "id": "1350625", "postDate": "06/15/2021 15:53:25", "content": "

I can offer my own version, just using albumentation. only my notepad is about writing coco json, but bbox change is happening there and checked. you need to make a dataset class with a transformation, then extract and write the coordinates where you need it.
\ncsv-to-coco-json-bbox-resize

", "rawMarkdown": "I can offer my own version, just using albumentation. only my notepad is about writing coco json, but bbox change is happening there and checked. you need to make a dataset class with a transformation, then extract and write the coordinates where you need it.\ncsv-to-coco-json-bbox-resize", "votes": null } ], "comments": [ { "id": 1348654, "author_name": "vbookshelf", "author_url": "", "post_date": "06/14/2021 07:07:00", "content": "

You can use the Albumentations package to resize both the images and bounding boxes. It's surprisingly simple. This is a link to a quick tutorial:
\nhttps://sheldonsebastian94.medium.com/resizing-image-and-bounding-boxes-for-object-detection-7b9d9463125a

", "votes": null, "replies": [] }, { "id": 1350625, "author_name": "zaakciiru", "author_url": "", "post_date": "06/15/2021 15:53:25", "content": "

I can offer my own version, just using albumentation. only my notepad is about writing coco json, but bbox change is happening there and checked. you need to make a dataset class with a transformation, then extract and write the coordinates where you need it.
\ncsv-to-coco-json-bbox-resize

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1348381": "The Original Dimensions of the Images are in the size of around 4000x4000.This is too much and resizing is a must . My plan is to convert into PNG formats and exploring various resolutions .The problem here is that we also need to deal with the boxes coordinates but just dividing them by the resolution is not the correct way .The boxes i made by this method were too off the real ones .Any ideas on how to deal with this situation are welcome .\nYou can view the difference in the image hosted in the link below.Resolution is 256x256 .\n![https://ibb.co/b1jSyFN]", "1348654": "You can use the Albumentations package to resize both the images and bounding boxes. It's surprisingly simple. This is a link to a quick tutorial:\nhttps://sheldonsebastian94.medium.com/resizing-image-and-bounding-boxes-for-object-detection-7b9d9463125a", "1350625": "I can offer my own version, just using albumentation. only my notepad is about writing coco json, but bbox change is happening there and checked. you need to make a dataset class with a transformation, then extract and write the coordinates where you need it.\ncsv-to-coco-json-bbox-resize" }, "source": "meta" }