Gentlemen,
\n\nWhats the difference or the connection between Metadata-initial-relationship-triplets and Metadata-relationship-triplets. (These are among the downloadable files in the challenge site)
", "messages": [ { "id": 358211, "postDate": "2018-07-17T18:02:28.507Z", "content": "Gentlemen,
\n\nWhats the difference or the connection between Metadata-initial-relationship-triplets and Metadata-relationship-triplets. (These are among the downloadable files in the challenge site)
", "rawMarkdown": "Gentlemen,\n\nWhats the difference or the connection between Metadata-initial-relationship-triplets and Metadata-relationship-triplets. (These are among the downloadable files in the challenge site)", "votes": 1 }, { "id": 361174, "postDate": "2018-07-24T00:44:20.283Z", "content": "Guys, one more question:
\n\nI downloaded validation data (41,620 imgs) from figure eight and I cross referenced it with the positive relationship triplets (374,768) and there isn't any relationship data for the validation set, it just has the bounding box details. So I am just confused, shouldn't the validation set have relationship description between Label1 and Label2 to validate? Or am I missing something
", "rawMarkdown": "Guys, one more question:\n\nI downloaded validation data (41,620 imgs) from figure eight and I cross referenced it with the positive relationship triplets (374,768) and there isn't any relationship data for the validation set, it just has the bounding box details. So I am just confused, shouldn't the validation set have relationship description between Label1 and Label2 to validate? Or am I missing something", "replies": [ { "id": 361619, "postDate": "2018-07-24T19:37:39.897Z", "content": "Yes, this is a bit confusing, sorry. The original OpenImages validation and test sets are not part of the OpenImage challenge. There are no relationships annotated for these sets and also the box annotations are less dense.
\n\nInstead, we only provide annotations for the training set of Open Images. within this set we have suggested which IDs to use as validation set. So it is advised to split the OpenImages training set into two parts according to these IDs.
", "rawMarkdown": "Yes, this is a bit confusing, sorry. The original OpenImages validation and test sets are not part of the OpenImage challenge. There are no relationships annotated for these sets and also the box annotations are less dense.\n\nInstead, we only provide annotations for the training set of Open Images. *within* this set we have suggested which IDs to use as validation set. So it is advised to split the OpenImages training set into two parts according to these IDs.", "votes": 1 }, { "id": 361621, "postDate": "2018-07-24T19:39:39.743Z", "content": "Perfect, thanks a lot Jasper
", "rawMarkdown": "Perfect, thanks a lot Jasper" } ] }, { "id": 360833, "postDate": "2018-07-23T09:54:45.227Z", "content": "The challenge is only about 329 triplets. This is summarized in Table 2 here: https://storage.googleapis.com/openimages/web/challenge.html
", "rawMarkdown": "The challenge is only about 329 triplets. This is summarized in Table 2 here: https://storage.googleapis.com/openimages/web/challenge.html ", "replies": [ { "id": 360914, "postDate": "2018-07-23T13:51:53.897Z", "content": "Thanks a lot @Jasper
", "rawMarkdown": "Thanks a lot @Jasper " }, { "id": 361172, "postDate": "2018-07-24T00:43:56.967Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 358518, "postDate": "2018-07-18T10:03:54.113Z", "content": "I have the same confusion. Does it mean that there would be 467 possible triplets in the dataset but the training set only contains 329 triplets?
", "rawMarkdown": "I have the same confusion. Does it mean that there would be 467 possible triplets in the dataset but the training set only contains 329 triplets?", "replies": [ { "id": 359660, "postDate": "2018-07-20T14:49:31.270Z", "content": "Based on reading the description on the open images website, I believe this to be correct. If I am reading it right, 467 possible triplets were identified but in the process of annotating the images only 329 were found in the dataset.
\n\n\n\n\n", "rawMarkdown": "Based on reading the description on the open images website, I believe this to be correct. If I am reading it right, 467 possible triplets were identified but in the process of annotating the images only 329 were found in the dataset.\n\n> The 329 of them that have at least one instance in the training set\n> form the final set of triplets for the Visual Relationships Detection\n> track\n\n\n[reference][1]\n\n\n [1]: https://storage.googleapis.com/openimages/web/challenge.html" }, { "id": 359663, "postDate": "2018-07-20T14:51:58.460Z", "content": "The 329 of them that have at least one instance in the training set\n form the final set of triplets for the Visual Relationships Detection\n track
\n
Might just try with both the values just to be sure. Thanks for the clarification Radek
", "rawMarkdown": "Might just try with both the values just to be sure. Thanks for the clarification Radek" } ] } ], "comments": [ { "id": 361174, "author_name": "Mukesh", "author_url": "", "post_date": "2018-07-24T00:44:20.283000", "content": "Guys, one more question:
\n\nI downloaded validation data (41,620 imgs) from figure eight and I cross referenced it with the positive relationship triplets (374,768) and there isn't any relationship data for the validation set, it just has the bounding box details. So I am just confused, shouldn't the validation set have relationship description between Label1 and Label2 to validate? Or am I missing something
", "votes": 0, "replies": [ { "id": 361619, "author_name": "Jasper Uijlings", "author_url": "", "post_date": "2018-07-24T19:37:39.897000", "content": "Yes, this is a bit confusing, sorry. The original OpenImages validation and test sets are not part of the OpenImage challenge. There are no relationships annotated for these sets and also the box annotations are less dense.
\n\nInstead, we only provide annotations for the training set of Open Images. within this set we have suggested which IDs to use as validation set. So it is advised to split the OpenImages training set into two parts according to these IDs.
", "votes": 1, "replies": [] }, { "id": 361621, "author_name": "Mukesh", "author_url": "", "post_date": "2018-07-24T19:39:39.743000", "content": "Perfect, thanks a lot Jasper
", "votes": 0, "replies": [] } ] }, { "id": 360833, "author_name": "Jasper Uijlings", "author_url": "", "post_date": "2018-07-23T09:54:45.227000", "content": "The challenge is only about 329 triplets. This is summarized in Table 2 here: https://storage.googleapis.com/openimages/web/challenge.html
", "votes": 0, "replies": [ { "id": 360914, "author_name": "Mukesh", "author_url": "", "post_date": "2018-07-23T13:51:53.897000", "content": "Thanks a lot @Jasper
", "votes": 0, "replies": [] }, { "id": 361172, "author_name": "", "author_url": "", "post_date": "2018-07-24T00:43:56.967000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 358518, "author_name": "Tianjian Meng", "author_url": "", "post_date": "2018-07-18T10:03:54.113000", "content": "I have the same confusion. Does it mean that there would be 467 possible triplets in the dataset but the training set only contains 329 triplets?
", "votes": 0, "replies": [ { "id": 359660, "author_name": "Radek Osmulski", "author_url": "", "post_date": "2018-07-20T14:49:31.270000", "content": "Based on reading the description on the open images website, I believe this to be correct. If I am reading it right, 467 possible triplets were identified but in the process of annotating the images only 329 were found in the dataset.
\n\n\n\n\n", "votes": 0, "replies": [] }, { "id": 359663, "author_name": "Mukesh", "author_url": "", "post_date": "2018-07-20T14:51:58.460000", "content": "The 329 of them that have at least one instance in the training set\n form the final set of triplets for the Visual Relationships Detection\n track
\n
Might just try with both the values just to be sure. Thanks for the clarification Radek
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "358211": "Gentlemen,\n\nWhats the difference or the connection between Metadata-initial-relationship-triplets and Metadata-relationship-triplets. (These are among the downloadable files in the challenge site)", "361174": "Guys, one more question:\n\nI downloaded validation data (41,620 imgs) from figure eight and I cross referenced it with the positive relationship triplets (374,768) and there isn't any relationship data for the validation set, it just has the bounding box details. So I am just confused, shouldn't the validation set have relationship description between Label1 and Label2 to validate? Or am I missing something", "360833": "The challenge is only about 329 triplets. This is summarized in Table 2 here: https://storage.googleapis.com/openimages/web/challenge.html ", "358518": "I have the same confusion. Does it mean that there would be 467 possible triplets in the dataset but the training set only contains 329 triplets?" } }