{ "id": 38571, "title": "What software to use to edit the masks in training sets to correct error", "url": "/competitions/carvana-image-masking-challenge/discussion/38571", "author_name": "XiaokangWang", "post_date": "2017-08-25T16:50:37.773000", "votes": 1, "comment_count": 8, "views": 0, "content": "

I saw errors in training set were pointed. I was thinking how to correct them?

", "messages": [ { "id": 216427, "postDate": "2017-08-25T18:57:05.767Z", "content": "

The thing is that there will be the same errors on the test set since it was a random split.

\n\n

So the competition is more about replicating the human error than getting a perfect mask.

", "rawMarkdown": "The thing is that there will be the same errors on the test set since it was a random split.\n\nSo the competition is more about replicating the human error than getting a perfect mask.", "votes": 1 }, { "id": 216406, "postDate": "2017-08-25T16:50:37.773Z", "content": "

I saw errors in training set were pointed. I was thinking how to correct them?

", "rawMarkdown": "I saw errors in training set were pointed. I was thinking how to correct them?", "votes": 1 }, { "id": 216422, "postDate": "2017-08-25T18:29:48.637Z", "content": "

It's prohibited by rules

", "rawMarkdown": "It's prohibited by rules", "votes": -3, "replies": [ { "id": 216426, "postDate": "2017-08-25T18:55:49.570Z", "content": "

Hum, I don't think so. You can do whatever you want with the training set. The winners of past competitions did do some manual editing on the training set.

", "rawMarkdown": "Hum, I don't think so. You can do whatever you want with the training set. The winners of past competitions did do some manual editing on the training set." }, { "id": 216439, "postDate": "2017-08-25T19:52:54.627Z", "content": "

If you may do whatever you want with the training set, may be you may do whatever you want even with the testing set?\nI mean that there no clarification from this point from Brian and Co.

", "rawMarkdown": "If you may do whatever you want with the training set, may be you may do whatever you want even with the testing set?\nI mean that there no clarification from this point from Brian and Co.", "votes": -2 }, { "id": 216592, "postDate": "2017-08-26T19:39:46.053Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 216713, "postDate": "2017-08-27T17:48:11.263Z", "content": "

As for the software to edit masks. You can use, for instance, ImageJ: load binary mask image, convert to ROI, edit ROI on top of color image of the car, then convert ROI back to binary mask. ImageJ is quite useful for a lot of other data exploration too.

", "rawMarkdown": "As for the software to edit masks. You can use, for instance, [ImageJ](https://imagej.nih.gov/ij/): load binary mask image, convert to ROI, edit ROI on top of color image of the car, then convert ROI back to binary mask. [ImageJ](https://imagej.nih.gov/ij/) is quite useful for a lot of other data exploration too.", "replies": [ { "id": 216791, "postDate": "2017-08-28T06:20:55.513Z", "content": "

Thanks Jurand. I will give it a try.

", "rawMarkdown": "Thanks Jurand. I will give it a try." } ] }, { "id": 216451, "postDate": "2017-08-25T20:33:15.293Z", "content": "

I agree with Charles Jansen. The test set might have similar errors but I think the majority of them have no error. So correcting errors in training set might help. Also, I 'd like to correct the very obvious errors in the training set.

", "rawMarkdown": "I agree with Charles Jansen. The test set might have similar errors but I think the majority of them have no error. So correcting errors in training set might help. Also, I 'd like to correct the very obvious errors in the training set." } ], "comments": [ { "id": 216427, "author_name": "Charles Jansen", "author_url": "", "post_date": "2017-08-25T18:57:05.767000", "content": "

The thing is that there will be the same errors on the test set since it was a random split.

\n\n

So the competition is more about replicating the human error than getting a perfect mask.

", "votes": 1, "replies": [] }, { "id": 216422, "author_name": "Integral", "author_url": "", "post_date": "2017-08-25T18:29:48.637000", "content": "

It's prohibited by rules

", "votes": -3, "replies": [ { "id": 216426, "author_name": "Charles Jansen", "author_url": "", "post_date": "2017-08-25T18:55:49.570000", "content": "

Hum, I don't think so. You can do whatever you want with the training set. The winners of past competitions did do some manual editing on the training set.

", "votes": 0, "replies": [] }, { "id": 216439, "author_name": "Integral", "author_url": "", "post_date": "2017-08-25T19:52:54.627000", "content": "

If you may do whatever you want with the training set, may be you may do whatever you want even with the testing set?\nI mean that there no clarification from this point from Brian and Co.

", "votes": -2, "replies": [] }, { "id": 216592, "author_name": "", "author_url": "", "post_date": "2017-08-26T19:39:46.053000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 216713, "author_name": "Jurand", "author_url": "", "post_date": "2017-08-27T17:48:11.263000", "content": "

As for the software to edit masks. You can use, for instance, ImageJ: load binary mask image, convert to ROI, edit ROI on top of color image of the car, then convert ROI back to binary mask. ImageJ is quite useful for a lot of other data exploration too.

", "votes": 0, "replies": [ { "id": 216791, "author_name": "XiaokangWang", "author_url": "", "post_date": "2017-08-28T06:20:55.513000", "content": "

Thanks Jurand. I will give it a try.

", "votes": 0, "replies": [] } ] }, { "id": 216451, "author_name": "XiaokangWang", "author_url": "", "post_date": "2017-08-25T20:33:15.293000", "content": "

I agree with Charles Jansen. The test set might have similar errors but I think the majority of them have no error. So correcting errors in training set might help. Also, I 'd like to correct the very obvious errors in the training set.

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "216427": "The thing is that there will be the same errors on the test set since it was a random split.\n\nSo the competition is more about replicating the human error than getting a perfect mask.", "216406": "I saw errors in training set were pointed. I was thinking how to correct them?", "216422": "It's prohibited by rules", "216713": "As for the software to edit masks. You can use, for instance, [ImageJ](https://imagej.nih.gov/ij/): load binary mask image, convert to ROI, edit ROI on top of color image of the car, then convert ROI back to binary mask. [ImageJ](https://imagej.nih.gov/ij/) is quite useful for a lot of other data exploration too.", "216451": "I agree with Charles Jansen. The test set might have similar errors but I think the majority of them have no error. So correcting errors in training set might help. Also, I 'd like to correct the very obvious errors in the training set." } }