There are some images really seeming containing target but not.
\nShould these images be dropped or keeping? (noise or mistake ?)Personally, I think these image would affect the performance of models, but I dont know its positive or negative effect.
right is the mask and left is the related image.
\n095bf7a1f_865_no.png
and
\n8242609fa_433_no.png
\nhttp://m.qpic.cn/psc?/V13TV6gC2cszdp/TmEUgtj9EK6.7V8ajmQrEHT0eCzAhzle6PzraCr.RAG7r5FwJyKTUQjZX*hrVLk1w56H8G02.kedY*DlHRJGZaABzIcbhl5Vq7EEZdoTFmk!/b&bo=SASQAUgEkAEDORw!&rf=viewer_4
and
\n\nand
\nwhich are similar to the target sample
☝️
\nUp!: lots of FN in image_id = 'afa5e8098'.
There are some images really seeming containing target but not.
\nShould these images be dropped or keeping? (noise or mistake ?)Personally, I think these image would affect the performance of models, but I dont know its positive or negative effect.
right is the mask and left is the related image.
\n095bf7a1f_865_no.png
and
\n8242609fa_433_no.png
\nhttp://m.qpic.cn/psc?/V13TV6gC2cszdp/TmEUgtj9EK6.7V8ajmQrEHT0eCzAhzle6PzraCr.RAG7r5FwJyKTUQjZX*hrVLk1w56H8G02.kedY*DlHRJGZaABzIcbhl5Vq7EEZdoTFmk!/b&bo=SASQAUgEkAEDORw!&rf=viewer_4
and
\n\nand
\nwhich are similar to the target sample
☝️
\nUp!: lots of FN in image_id = 'afa5e8098'.
You should drop or edit csv to include these masks. I think some people already reported missing labels in the training set.
", "rawMarkdown": "You should drop or edit csv to include these masks. I think some people already reported missing labels in the training set.", "votes": 1, "replies": [ { "id": 1272350, "postDate": "2021-04-13T12:13:21.903Z", "content": "I will have a try
", "rawMarkdown": "I will have a try" } ] }, { "id": 1272749, "postDate": "2021-04-13T18:14:10.447Z", "content": "Are you sure your code is generating the mask correctly? Those both look like glomeruli to me. Can you share where on the image these are?
\nAlso if you want to embed images, you need to: 
095bf7a1f_865_no.png. Using 1024*1024 tiles without overlap.
", "rawMarkdown": "095bf7a1f_865_no.png. Using 1024*1024 tiles without overlap.", "votes": 2 }, { "id": 1272991, "postDate": "2021-04-14T01:46:22.190Z", "content": "thanks for your help!
", "rawMarkdown": "thanks for your help!", "votes": 1 }, { "id": 1272995, "postDate": "2021-04-14T01:52:29.973Z", "content": "I've post the position of images. Could you have check with that?
", "rawMarkdown": "I've post the position of images. Could you have check with that?" }, { "id": 1273030, "postDate": "2021-04-14T03:03:18.327Z", "content": "Yup, I think those are false negatives.
", "rawMarkdown": "Yup, I think those are false negatives." } ] }, { "id": 1279302, "postDate": "2021-04-20T19:14:14.983Z", "content": "By some reason i can't see the images, but it got my attention, how did you identified those? I'm not addressing any kind of problem like that on my code
", "rawMarkdown": "By some reason i can't see the images, but it got my attention, how did you identified those? I'm not addressing any kind of problem like that on my code", "replies": [ { "id": 1281735, "postDate": "2021-04-23T08:52:19.457Z", "content": "I used a simple classifier finding them.
", "rawMarkdown": "I used a simple classifier finding them." } ] }, { "id": 1278983, "postDate": "2021-04-20T13:28:07.990Z", "content": "There is a possibility that we try to fit an automatically generated ground truth followed by a visual assertion by an expert. This kind of data generating process has some drawbacks.
", "rawMarkdown": "There is a possibility that we try to fit an automatically generated ground truth followed by a visual assertion by an expert. This kind of data generating process has some drawbacks." }, { "id": 1275243, "postDate": "2021-04-16T05:50:27.490Z", "content": "One can only hope the private test is annotated consistently. Otherwise the most accurate models will not win this competition; the models that match human levels of performance will win.
", "rawMarkdown": "One can only hope the private test is annotated consistently. Otherwise the most accurate models will not win this competition; the models that match human levels of performance will win.", "replies": [ { "id": 1275299, "postDate": "2021-04-16T06:59:38.303Z", "content": "Yup. I'm waiting for holders' response but they are still keeping in slience.
", "rawMarkdown": "Yup. I'm waiting for holders' response but they are still keeping in slience." } ] }, { "id": 1271997, "postDate": "2021-04-13T06:30:25.480Z", "content": "The second is a glomeruli for sure. The first, harder to confirm, but probably is too.
", "rawMarkdown": "The second is a glomeruli for sure. The first, harder to confirm, but probably is too. ", "replies": [ { "id": 1272352, "postDate": "2021-04-13T12:14:32.470Z", "content": "I think so. I will upload more confusing images
", "rawMarkdown": "I think so. I will upload more confusing images" } ] }, { "id": 1271907, "postDate": "2021-04-13T03:42:13.430Z", "content": "those don't look like glomeruli to me
", "rawMarkdown": "those don't look like glomeruli to me", "replies": [ { "id": 1272351, "postDate": "2021-04-13T12:14:10.073Z", "content": "I have found other images which seem to be noises too. I will add some more image links
", "rawMarkdown": "I have found other images which seem to be noises too. I will add some more image links" } ] }, { "id": 1271821, "postDate": "2021-04-13T00:08:04.987Z", "content": "Im curious about the amount of false negative.
", "rawMarkdown": "Im curious about the amount of false negative. " }, { "id": 1271200, "postDate": "2021-04-12T11:51:52.180Z", "content": "I think is mistake, you need to drop the mask.😄
", "rawMarkdown": "I think is mistake, you need to drop the mask.😄" }, { "id": 1270946, "postDate": "2021-04-12T07:16:48.843Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1271048, "author_name": "SHih Chieh Lai", "author_url": "", "post_date": "2021-04-12T08:41:06.540000", "content": "You should drop or edit csv to include these masks. I think some people already reported missing labels in the training set.
", "votes": 1, "replies": [ { "id": 1272350, "author_name": "朴大福", "author_url": "", "post_date": "2021-04-13T12:13:21.903000", "content": "I will have a try
", "votes": 0, "replies": [] } ] }, { "id": 1272749, "author_name": "Erik Dali", "author_url": "", "post_date": "2021-04-13T18:14:10.447000", "content": "Are you sure your code is generating the mask correctly? Those both look like glomeruli to me. Can you share where on the image these are?
\nAlso if you want to embed images, you need to: 
095bf7a1f_865_no.png. Using 1024*1024 tiles without overlap.
", "votes": 2, "replies": [] }, { "id": 1272991, "author_name": "朴大福", "author_url": "", "post_date": "2021-04-14T01:46:22.190000", "content": "thanks for your help!
", "votes": 1, "replies": [] }, { "id": 1272995, "author_name": "朴大福", "author_url": "", "post_date": "2021-04-14T01:52:29.973000", "content": "I've post the position of images. Could you have check with that?
", "votes": 0, "replies": [] }, { "id": 1273030, "author_name": "Erik Dali", "author_url": "", "post_date": "2021-04-14T03:03:18.327000", "content": "Yup, I think those are false negatives.
", "votes": 0, "replies": [] } ] }, { "id": 1279302, "author_name": "victorasso", "author_url": "", "post_date": "2021-04-20T19:14:14.983000", "content": "By some reason i can't see the images, but it got my attention, how did you identified those? I'm not addressing any kind of problem like that on my code
", "votes": 0, "replies": [ { "id": 1281735, "author_name": "朴大福", "author_url": "", "post_date": "2021-04-23T08:52:19.457000", "content": "I used a simple classifier finding them.
", "votes": 0, "replies": [] } ] }, { "id": 1278983, "author_name": "Alexandre Cadrin-Chênevert", "author_url": "", "post_date": "2021-04-20T13:28:07.990000", "content": "There is a possibility that we try to fit an automatically generated ground truth followed by a visual assertion by an expert. This kind of data generating process has some drawbacks.
", "votes": 0, "replies": [] }, { "id": 1275243, "author_name": "Erik Dali", "author_url": "", "post_date": "2021-04-16T05:50:27.490000", "content": "One can only hope the private test is annotated consistently. Otherwise the most accurate models will not win this competition; the models that match human levels of performance will win.
", "votes": 0, "replies": [ { "id": 1275299, "author_name": "朴大福", "author_url": "", "post_date": "2021-04-16T06:59:38.303000", "content": "Yup. I'm waiting for holders' response but they are still keeping in slience.
", "votes": 0, "replies": [] } ] }, { "id": 1271997, "author_name": "rpsantosa_kaggle", "author_url": "", "post_date": "2021-04-13T06:30:25.480000", "content": "The second is a glomeruli for sure. The first, harder to confirm, but probably is too.
", "votes": 0, "replies": [ { "id": 1272352, "author_name": "朴大福", "author_url": "", "post_date": "2021-04-13T12:14:32.470000", "content": "I think so. I will upload more confusing images
", "votes": 0, "replies": [] } ] }, { "id": 1271907, "author_name": "andras", "author_url": "", "post_date": "2021-04-13T03:42:13.430000", "content": "those don't look like glomeruli to me
", "votes": 0, "replies": [ { "id": 1272351, "author_name": "朴大福", "author_url": "", "post_date": "2021-04-13T12:14:10.073000", "content": "I have found other images which seem to be noises too. I will add some more image links
", "votes": 0, "replies": [] } ] }, { "id": 1271821, "author_name": "rpsantosa_kaggle", "author_url": "", "post_date": "2021-04-13T00:08:04.987000", "content": "Im curious about the amount of false negative.
", "votes": 0, "replies": [] }, { "id": 1271200, "author_name": "Jason Huang", "author_url": "", "post_date": "2021-04-12T11:51:52.180000", "content": "I think is mistake, you need to drop the mask.😄
", "votes": 0, "replies": [] }, { "id": 1270946, "author_name": "", "author_url": "", "post_date": "2021-04-12T07:16:48.843000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1270936": "There are some images really seeming containing target but not. \nShould these images be dropped or keeping? (noise or mistake ?)Personally, I think these image would affect the performance of models, but I dont know its positive or negative effect.\n\nright is the mask and left is the related image.\n095bf7a1f_865_no.png\n\nhttp://m.qpic.cn/psc?/V13TV6gC2cszdp/TmEUgtj9EK6.7V8ajmQrEOZRFDYkFJ8xVjW1fjFg5SsfwCKPVfazKlS8ZoBL*KEJtMm3HGnvw8KVOKRGVR*C57zdP9bvOz4Jia8GIhwfegQ!/b&bo=LwSaAS8EmgEDORw!&rf=viewer_4 \n\nand\n8242609fa_433_no.png\nhttp://m.qpic.cn/psc?/V13TV6gC2cszdp/TmEUgtj9EK6.7V8ajmQrEHT0eCzAhzle6PzraCr.RAG7r5FwJyKTUQjZX*hrVLk1w56H8G02.kedY*DlHRJGZaABzIcbhl5Vq7EEZdoTFmk!/b&bo=SASQAUgEkAEDORw!&rf=viewer_4\n\nand\n\nhttp://m.qpic.cn/psc?/V13TV6gC2cszdp/TmEUgtj9EK6.7V8ajmQrEKRYah8poKgdKjK4JSNa1DQA3rI53coqcc8py7j9LDL6tR3YNloxrsGW3qLRt6VrP4KXi*0.fKUw9lAxXpa*LCg!/b&bo=UwSQAVMEkAEDKQw!&rf=viewer_4\n\nand\nwhich are similar to the target sample\n\nhttp://m.qpic.cn/psc?/V13TV6gC2cszdp/TmEUgtj9EK6.7V8ajmQrELavQSf4.sRTakToB3Rqebpe73Id5KQZ7csdnBaXEhObJnFrOPsbel6dmSD4MzCeYYXvsdoTQoYJ2pewFS2zrjc!/b&bo=BgXFAQYFxQEDSWw!&rf=viewer_4\n\n☝️\n**Up!**: lots of FN in image_id = 'afa5e8098'.", "1271048": "You should drop or edit csv to include these masks. I think some people already reported missing labels in the training set.", "1272749": "Are you sure your code is generating the mask correctly? Those both look like glomeruli to me. Can you share where on the image these are?\n\nAlso if you want to embed images, you need to: ``", "1279302": "By some reason i can't see the images, but it got my attention, how did you identified those? I'm not addressing any kind of problem like that on my code", "1278983": "There is a possibility that we try to fit an automatically generated ground truth followed by a visual assertion by an expert. This kind of data generating process has some drawbacks.", "1275243": "One can only hope the private test is annotated consistently. Otherwise the most accurate models will not win this competition; the models that match human levels of performance will win.", "1271997": "The second is a glomeruli for sure. The first, harder to confirm, but probably is too. ", "1271907": "those don't look like glomeruli to me", "1271821": "Im curious about the amount of false negative. ", "1271200": "I think is mistake, you need to drop the mask.😄", "1270946": "" } }