After days of \"Submission CSV Not Found\" I finally figured out why I was getting errors.\nIs this printed somewhere that I missed??
", "messages": [ { "id": 841106, "postDate": "2020-05-10T15:48:02.477Z", "content": "After days of \"Submission CSV Not Found\" I finally figured out why I was getting errors.\nIs this printed somewhere that I missed??
", "rawMarkdown": "After days of \"Submission CSV Not Found\" I finally figured out why I was getting errors.\nIs this printed somewhere that I missed??", "votes": 10 }, { "id": 845533, "postDate": "2020-05-13T09:07:37.607Z", "content": "Thank you for raising these issue. Normally images are all 1024px X 1024px, but we will investigate the test set if there is any problem related to your issue.
", "rawMarkdown": "Thank you for raising these issue. Normally images are all 1024px X 1024px, but we will investigate the test set if there is any problem related to your issue.", "votes": 3, "replies": [ { "id": 850995, "postDate": "2020-05-17T08:42:37.787Z", "content": "@bendvd hi, can you ensure the image size consistency in both training set and test set.
", "rawMarkdown": "@bendvd hi, can you ensure the image size consistency in both training set and test set. " }, { "id": 897188, "postDate": "2020-06-22T17:23:28.183Z", "content": "@bendvd any update on this? We don't know whether we should be investing time in guarding against this.
", "rawMarkdown": "@bendvd any update on this? We don't know whether we should be investing time in guarding against this." }, { "id": 897208, "postDate": "2020-06-22T17:49:49.310Z", "content": "@alexandersoare its pathetic, they are a very inactive host ☹️
", "rawMarkdown": "@alexandersoare its pathetic, they are a very inactive host ☹️ " }, { "id": 897844, "postDate": "2020-06-23T06:15:46.583Z", "content": "one of the test set is still 1200x1200 px and not 1024. However by resizing the image to 1024 should do the trick. As mentioned in https://www.kaggle.com/c/global-wheat-detection/discussion/159585 , we are sorry for the lack of feedback during the last 3/4 weeks and hope to solve these bugs. There are lot of technical difficulties to consider before updating the dataset, but we are working on a satisfactory solution for everybody.
", "rawMarkdown": "one of the test set is still 1200x1200 px and not 1024. However by resizing the image to 1024 should do the trick. As mentioned in https://www.kaggle.com/c/global-wheat-detection/discussion/159585 , we are sorry for the lack of feedback during the last 3/4 weeks and hope to solve these bugs. There are lot of technical difficulties to consider before updating the dataset, but we are working on a satisfactory solution for everybody.", "votes": 6 }, { "id": 897859, "postDate": "2020-06-23T06:36:58.600Z", "content": "Good to hear the problem is only one image. Shouldn't be a problem at all :)
", "rawMarkdown": "Good to hear the problem is only one image. Shouldn't be a problem at all :)" }, { "id": 897886, "postDate": "2020-06-23T06:58:49.917Z", "content": "@bendvd should we be rescaling our predicted bounding boxes to match whatever the image size is? I actually tested this once and didn't get a different LB score.
", "rawMarkdown": "@bendvd should we be rescaling our predicted bounding boxes to match whatever the image size is? I actually tested this once and didn't get a different LB score.", "votes": 2 }, { "id": 899254, "postDate": "2020-06-24T05:37:39.837Z", "content": "it is a problem on 20 images to be fair, so not the majority of the test set. However we will try to solve this issue in the coming days.
", "rawMarkdown": "it is a problem on 20 images to be fair, so not the majority of the test set. However we will try to solve this issue in the coming days.", "votes": 1 }, { "id": 901848, "postDate": "2020-06-25T18:28:29.880Z", "content": "Any updates on this issue? As those image may be in private set so we won't see much diff for now. However, it may affect final results as there are limited images on private. So which way should we do @bendvd ? \n1) rescale all images during inference to 1024 regardless actual size, and generate box strictly to 1024 \n2) resize all boxes to actual image size during inference
", "rawMarkdown": "Any updates on this issue? As those image may be in private set so we won't see much diff for now. However, it may affect final results as there are limited images on private. So which way should we do @bendvd ? \n1) rescale all images during inference to 1024 regardless actual size, and generate box strictly to 1024 \n2) resize all boxes to actual image size during inference \n", "votes": 2 }, { "id": 902523, "postDate": "2020-06-26T07:45:04.843Z", "content": "Sorry @weimin , we still need a couple of days to bring a solution. I know the waiting is frustrating and I'm also sorry !
", "rawMarkdown": "Sorry @weimin , we still need a couple of days to bring a solution. I know the waiting is frustrating and I'm also sorry !" }, { "id": 913517, "postDate": "2020-07-03T08:55:59.033Z", "content": "It will be rescale to 1024 x 1024 for the final leaderboard.
", "rawMarkdown": "It will be rescale to 1024 x 1024 for the final leaderboard. ", "votes": 2 }, { "id": 918678, "postDate": "2020-07-07T11:53:59.540Z", "content": "@bendvd to make further clarification, does this mean that all the solutions will be rerun once again?
", "rawMarkdown": "@bendvd to make further clarification, does this mean that all the solutions will be rerun once again? " }, { "id": 919077, "postDate": "2020-07-07T17:39:02.250Z", "content": "No it doesn’t. If you look at the competition update, they make it clear that ONLY the 38% of private test data that is used for final scoring will be corrected.
", "rawMarkdown": "No it doesn’t. If you look at the competition update, they make it clear that ONLY the 38% of private test data that is used for final scoring will be corrected.", "votes": 1 } ] }, { "id": 855024, "postDate": "2020-05-20T14:12:28.127Z", "content": "Thanks for pointing this out! You are correct, I've tried random cutout and my code was breaking because of this while performing pseudo labeling.
", "rawMarkdown": "Thanks for pointing this out! You are correct, I've tried random cutout and my code was breaking because of this while performing pseudo labeling. " }, { "id": 841448, "postDate": "2020-05-10T19:48:18.827Z", "content": "Well thanks for bringing this up, I just confirmed that this is the case with my kernel.
", "rawMarkdown": "Well thanks for bringing this up, I just confirmed that this is the case with my kernel." }, { "id": 841156, "postDate": "2020-05-10T16:39:47.843Z", "content": "Thanks for sharing. Are you sure about this ? Because there is not such issue in the training set which is supposed to represent 73% of the whole set. How did you find out ?
", "rawMarkdown": "Thanks for sharing. Are you sure about this ? Because there is not such issue in the training set which is supposed to represent 73% of the whole set. How did you find out ?", "replies": [ { "id": 841185, "postDate": "2020-05-10T16:57:38.150Z", "content": "My model must take 1024x1024 else it errors, all my submissions failed early within 30 seconds-1 minute (changes because the order the images are in changes). Then when I added a resize from looking at other people's inference kernals, it worked (took 6~ minutes to submit). I then removed the resize without changing anything else and it errored out quickly again.\nAlso as stated in other discussions the test set is mostly completely different images from unique locations.
", "rawMarkdown": "My model must take 1024x1024 else it errors, all my submissions failed early within 30 seconds-1 minute (changes because the order the images are in changes). Then when I added a resize from looking at other people's inference kernals, it worked (took 6~ minutes to submit). I then removed the resize without changing anything else and it errored out quickly again.\nAlso as stated in other discussions the test set is mostly completely different images from unique locations.", "votes": 5 }, { "id": 841191, "postDate": "2020-05-10T17:00:10.350Z", "content": "Thanks for your reply !
", "rawMarkdown": "Thanks for your reply !" } ] }, { "id": 841700, "postDate": "2020-05-11T01:16:13.760Z", "rawMarkdown": "", "votes": -1, "isDeleted": true, "replies": [ { "id": 841721, "postDate": "2020-05-11T01:30:10.870Z", "content": "That is only a very small subset of the actual test set.
", "rawMarkdown": "That is only a very small subset of the actual test set.", "votes": 2 } ] } ], "comments": [ { "id": 845533, "author_name": "Etienne David", "author_url": "", "post_date": "2020-05-13T09:07:37.607000", "content": "Thank you for raising these issue. Normally images are all 1024px X 1024px, but we will investigate the test set if there is any problem related to your issue.
", "votes": 3, "replies": [ { "id": 850995, "author_name": "Innat", "author_url": "", "post_date": "2020-05-17T08:42:37.787000", "content": "@bendvd hi, can you ensure the image size consistency in both training set and test set.
", "votes": 0, "replies": [] }, { "id": 897188, "author_name": "Alexander Soare", "author_url": "", "post_date": "2020-06-22T17:23:28.183000", "content": "@bendvd any update on this? We don't know whether we should be investing time in guarding against this.
", "votes": 0, "replies": [] }, { "id": 897208, "author_name": "Innat", "author_url": "", "post_date": "2020-06-22T17:49:49.310000", "content": "@alexandersoare its pathetic, they are a very inactive host ☹️
", "votes": 0, "replies": [] }, { "id": 897844, "author_name": "Etienne David", "author_url": "", "post_date": "2020-06-23T06:15:46.583000", "content": "one of the test set is still 1200x1200 px and not 1024. However by resizing the image to 1024 should do the trick. As mentioned in https://www.kaggle.com/c/global-wheat-detection/discussion/159585 , we are sorry for the lack of feedback during the last 3/4 weeks and hope to solve these bugs. There are lot of technical difficulties to consider before updating the dataset, but we are working on a satisfactory solution for everybody.
", "votes": 6, "replies": [] }, { "id": 897859, "author_name": "Matt", "author_url": "", "post_date": "2020-06-23T06:36:58.600000", "content": "Good to hear the problem is only one image. Shouldn't be a problem at all :)
", "votes": 0, "replies": [] }, { "id": 897886, "author_name": "Alexander Soare", "author_url": "", "post_date": "2020-06-23T06:58:49.917000", "content": "@bendvd should we be rescaling our predicted bounding boxes to match whatever the image size is? I actually tested this once and didn't get a different LB score.
", "votes": 2, "replies": [] }, { "id": 899254, "author_name": "Etienne David", "author_url": "", "post_date": "2020-06-24T05:37:39.837000", "content": "it is a problem on 20 images to be fair, so not the majority of the test set. However we will try to solve this issue in the coming days.
", "votes": 1, "replies": [] }, { "id": 901848, "author_name": "Weimin Wang", "author_url": "", "post_date": "2020-06-25T18:28:29.880000", "content": "Any updates on this issue? As those image may be in private set so we won't see much diff for now. However, it may affect final results as there are limited images on private. So which way should we do @bendvd ? \n1) rescale all images during inference to 1024 regardless actual size, and generate box strictly to 1024 \n2) resize all boxes to actual image size during inference
", "votes": 2, "replies": [] }, { "id": 902523, "author_name": "Etienne David", "author_url": "", "post_date": "2020-06-26T07:45:04.843000", "content": "Sorry @weimin , we still need a couple of days to bring a solution. I know the waiting is frustrating and I'm also sorry !
", "votes": 0, "replies": [] }, { "id": 913517, "author_name": "Etienne David", "author_url": "", "post_date": "2020-07-03T08:55:59.033000", "content": "It will be rescale to 1024 x 1024 for the final leaderboard.
", "votes": 2, "replies": [] }, { "id": 918678, "author_name": "Seva Konjahhin", "author_url": "", "post_date": "2020-07-07T11:53:59.540000", "content": "@bendvd to make further clarification, does this mean that all the solutions will be rerun once again?
", "votes": 0, "replies": [] }, { "id": 919077, "author_name": "Matt", "author_url": "", "post_date": "2020-07-07T17:39:02.250000", "content": "No it doesn’t. If you look at the competition update, they make it clear that ONLY the 38% of private test data that is used for final scoring will be corrected.
", "votes": 1, "replies": [] } ] }, { "id": 855024, "author_name": "Kaushal Shah", "author_url": "", "post_date": "2020-05-20T14:12:28.127000", "content": "Thanks for pointing this out! You are correct, I've tried random cutout and my code was breaking because of this while performing pseudo labeling.
", "votes": 0, "replies": [] }, { "id": 841448, "author_name": "Matt", "author_url": "", "post_date": "2020-05-10T19:48:18.827000", "content": "Well thanks for bringing this up, I just confirmed that this is the case with my kernel.
", "votes": 0, "replies": [] }, { "id": 841156, "author_name": "Jay L", "author_url": "", "post_date": "2020-05-10T16:39:47.843000", "content": "Thanks for sharing. Are you sure about this ? Because there is not such issue in the training set which is supposed to represent 73% of the whole set. How did you find out ?
", "votes": 0, "replies": [ { "id": 841185, "author_name": "GreatGameDota", "author_url": "", "post_date": "2020-05-10T16:57:38.150000", "content": "My model must take 1024x1024 else it errors, all my submissions failed early within 30 seconds-1 minute (changes because the order the images are in changes). Then when I added a resize from looking at other people's inference kernals, it worked (took 6~ minutes to submit). I then removed the resize without changing anything else and it errored out quickly again.\nAlso as stated in other discussions the test set is mostly completely different images from unique locations.
", "votes": 5, "replies": [] }, { "id": 841191, "author_name": "Jay L", "author_url": "", "post_date": "2020-05-10T17:00:10.350000", "content": "Thanks for your reply !
", "votes": 0, "replies": [] } ] }, { "id": 841700, "author_name": "", "author_url": "", "post_date": "2020-05-11T01:16:13.760000", "content": "", "votes": -1, "replies": [ { "id": 841721, "author_name": "GreatGameDota", "author_url": "", "post_date": "2020-05-11T01:30:10.870000", "content": "That is only a very small subset of the actual test set.
", "votes": 2, "replies": [] } ] } ], "raw_markdown_by_id": { "841106": "After days of \"Submission CSV Not Found\" I finally figured out why I was getting errors.\nIs this printed somewhere that I missed??", "845533": "Thank you for raising these issue. Normally images are all 1024px X 1024px, but we will investigate the test set if there is any problem related to your issue.", "855024": "Thanks for pointing this out! You are correct, I've tried random cutout and my code was breaking because of this while performing pseudo labeling. ", "841448": "Well thanks for bringing this up, I just confirmed that this is the case with my kernel.", "841156": "Thanks for sharing. Are you sure about this ? Because there is not such issue in the training set which is supposed to represent 73% of the whole set. How did you find out ?", "841700": "" } }