In the specs on Github, you mentioned that we can't use annotations for the test set. Does that mean that \n(a) we can't generate the bounding box data, and then load it in our model, or\n(b) we can't use a detection model such as FRCNN or MobileNet inside our kernel to generate bounding boxes?
", "messages": [ { "id": "502416", "postDate": "03/28/2019 14:57:57", "content": "In the specs on Github, you mentioned that we can't use annotations for the test set. Does that mean that \n(a) we can't generate the bounding box data, and then load it in our model, or\n(b) we can't use a detection model such as FRCNN or MobileNet inside our kernel to generate bounding boxes?
", "rawMarkdown": "In the specs on Github, you mentioned that we can't use annotations for the test set. Does that mean that \n(a) we can't generate the bounding box data, and then load it in our model, or\n(b) we can't use a detection model such as FRCNN or MobileNet inside our kernel to generate bounding boxes?", "votes": null }, { "id": "502510", "postDate": "03/28/2019 17:04:01", "content": "I mean that you cannot explicitly collect ground truth bounding boxes or species identifications for the test set. You are 100% allowed to run any detector you would like over the test set, as long as that detector is either the one we have provided (and we have released the detection results for that, so you should be able to use them directly without running inference) or a detector trained on only the data allowed for the competition (Imagenet, COCO, CCT, or iNaturalist 2017-2019)
", "rawMarkdown": "I mean that you cannot explicitly collect ground truth bounding boxes or species identifications for the test set. You are 100% allowed to run any detector you would like over the test set, as long as that detector is either the one we have provided (and we have released the detection results for that, so you should be able to use them directly without running inference) or a detector trained on only the data allowed for the competition (Imagenet, COCO, CCT, or iNaturalist 2017-2019)", "votes": null }, { "id": "502524", "postDate": "03/28/2019 17:24:32", "content": "\n\n\n(and we have released the detection results for that, so you should be able to use them directly without running inference)
\n
Thank you for the quick answer! I was under the impression that the annotation results (bounding boxes) were only available for the data in the training sets, but not in the test sets. Are they both available on Github?
", "rawMarkdown": ">(and we have released the detection results for that, so you should be able to use them directly without running inference)\n\nThank you for the quick answer! I was under the impression that the annotation results (bounding boxes) were only available for the data in the training sets, but not in the test sets. Are they both available on Github?", "votes": null }, { "id": "502531", "postDate": "03/28/2019 17:35:51", "content": "Yep! They're all available on Github. But they probably aren't perfect! We don't have ground truth for the test data, so I don't have any metrics for the detection results.
", "rawMarkdown": "Yep! They're all available on Github. But they probably aren't perfect! We don't have ground truth for the test data, so I don't have any metrics for the detection results.", "votes": null }, { "id": "502533", "postDate": "03/28/2019 17:41:21", "content": "Sounds good! I probably missed it in this case. I'll take a closer look and try to make use of it. Thank you!
", "rawMarkdown": "Sounds good! I probably missed it in this case. I'll take a closer look and try to make use of it. Thank you!", "votes": null }, { "id": "502537", "postDate": "03/28/2019 17:46:11", "content": "NP :)
", "rawMarkdown": "NP :)", "votes": null }, { "id": "509355", "postDate": "04/07/2019 17:45:57", "content": "Where you able to find the annotation results for the test sets? In the file provided on the GitHub (https://wildcamdrop.blob.core.windows.net/wildcamdropcontainer/Detection_Results.tar.gz) the test annotations file is empty. Is it available somewhere else?
", "rawMarkdown": "Where you able to find the annotation results for the test sets? In the file provided on the GitHub (https://wildcamdrop.blob.core.windows.net/wildcamdropcontainer/Detection_Results.tar.gz) the test annotations file is empty. Is it available somewhere else?", "votes": null }, { "id": "510107", "postDate": "04/08/2019 16:56:38", "content": "I will look into this, sounds like I uploaded the wrong file.
", "rawMarkdown": "I will look into this, sounds like I uploaded the wrong file.", "votes": null }, { "id": "510210", "postDate": "04/08/2019 20:24:21", "content": "I did. The ''Detection_Results'' zipped folder has now been updated to include the IDFG results, split across two files: 'IDFG_Detection_Results_1.p' and 'IDFG_Detection_Results_2.p'
\n\nSo sorry for the confusion!
", "rawMarkdown": "I did. The ''Detection\\_Results'' zipped folder has now been updated to include the IDFG results, split across two files: 'IDFG\\_Detection\\_Results\\_1.p' and 'IDFG\\_Detection\\_Results\\_2.p'\n\nSo sorry for the confusion!", "votes": null }, { "id": "510241", "postDate": "04/08/2019 21:31:00", "content": "Thank you!
", "rawMarkdown": "Thank you!", "votes": null }, { "id": "519043", "postDate": "04/18/2019 09:17:08", "content": "As for the classification, is the bounding box data helpful?
", "rawMarkdown": "As for the classification, is the bounding box data helpful?", "votes": null }, { "id": "542255", "postDate": "06/03/2019 16:09:32", "content": "@sbeery \nI am sorry, I am confusing about the Detection_Results. \nThere are only 5 files in Detection_Results:\n['iNat_Idaho_Detection_Results.p', 'IDFG_Detection_Results_1.p', 'IDFG_Detection_Results_2.p', 'CCT_Detection_Results_2.p', 'CCT_Detection_Results_1.p']\n1. It seems to be, Detection_Results not include any test_images. Which is detection results for test_images?\n2. It seems to be, these are Bbox for full-size images, but I do not Know shapes for full-size images. So, How can I transfor the bbox to small-size images?\nThankyou
", "rawMarkdown": "sbeery \nI am sorry, I am confusing about the Detection_Results. \nThere are only 5 files in Detection_Results:\n['iNat_Idaho_Detection_Results.p', 'IDFG_Detection_Results_1.p', 'IDFG_Detection_Results_2.p', 'CCT_Detection_Results_2.p', 'CCT_Detection_Results_1.p']\n1. It seems to be, Detection_Results not include any test_images. Which is detection results for test_images?\n2. It seems to be, these are Bbox for full-size images, but I do not Know shapes for full-size images. So, How can I transfor the bbox to small-size images?\nThankyou", "votes": null }, { "id": "542421", "postDate": "06/03/2019 22:08:59", "content": "IDFGDetectionResults1 and 2 contain the results for the test images. The full image sizes are in the Annotation and Information files for each dataset.
", "rawMarkdown": "IDFGDetectionResults1 and 2 contain the results for the test images. The full image sizes are in the Annotation and Information files for each dataset.", "votes": null } ], "comments": [ { "id": 502510, "author_name": "sbeery", "author_url": "", "post_date": "03/28/2019 17:04:01", "content": "I mean that you cannot explicitly collect ground truth bounding boxes or species identifications for the test set. You are 100% allowed to run any detector you would like over the test set, as long as that detector is either the one we have provided (and we have released the detection results for that, so you should be able to use them directly without running inference) or a detector trained on only the data allowed for the competition (Imagenet, COCO, CCT, or iNaturalist 2017-2019)
", "votes": null, "replies": [ { "id": 502524, "author_name": "xhlulu", "author_url": "", "post_date": "03/28/2019 17:24:32", "content": "\n\n\n(and we have released the detection results for that, so you should be able to use them directly without running inference)
\n
Thank you for the quick answer! I was under the impression that the annotation results (bounding boxes) were only available for the data in the training sets, but not in the test sets. Are they both available on Github?
", "votes": null, "replies": [] }, { "id": 502531, "author_name": "sbeery", "author_url": "", "post_date": "03/28/2019 17:35:51", "content": "Yep! They're all available on Github. But they probably aren't perfect! We don't have ground truth for the test data, so I don't have any metrics for the detection results.
", "votes": null, "replies": [] }, { "id": 502533, "author_name": "xhlulu", "author_url": "", "post_date": "03/28/2019 17:41:21", "content": "Sounds good! I probably missed it in this case. I'll take a closer look and try to make use of it. Thank you!
", "votes": null, "replies": [] }, { "id": 502537, "author_name": "sbeery", "author_url": "", "post_date": "03/28/2019 17:46:11", "content": "NP :)
", "votes": null, "replies": [] }, { "id": 509355, "author_name": "davidparamo", "author_url": "", "post_date": "04/07/2019 17:45:57", "content": "Where you able to find the annotation results for the test sets? In the file provided on the GitHub (https://wildcamdrop.blob.core.windows.net/wildcamdropcontainer/Detection_Results.tar.gz) the test annotations file is empty. Is it available somewhere else?
", "votes": null, "replies": [] }, { "id": 510107, "author_name": "sbeery", "author_url": "", "post_date": "04/08/2019 16:56:38", "content": "I will look into this, sounds like I uploaded the wrong file.
", "votes": null, "replies": [] }, { "id": 510210, "author_name": "sbeery", "author_url": "", "post_date": "04/08/2019 20:24:21", "content": "I did. The ''Detection_Results'' zipped folder has now been updated to include the IDFG results, split across two files: 'IDFG_Detection_Results_1.p' and 'IDFG_Detection_Results_2.p'
\n\nSo sorry for the confusion!
", "votes": null, "replies": [] }, { "id": 510241, "author_name": "davidparamo", "author_url": "", "post_date": "04/08/2019 21:31:00", "content": "Thank you!
", "votes": null, "replies": [] } ] }, { "id": 519043, "author_name": "ganweifa123", "author_url": "", "post_date": "04/18/2019 09:17:08", "content": "As for the classification, is the bounding box data helpful?
", "votes": null, "replies": [] }, { "id": 542255, "author_name": "walleclipse", "author_url": "", "post_date": "06/03/2019 16:09:32", "content": "@sbeery \nI am sorry, I am confusing about the Detection_Results. \nThere are only 5 files in Detection_Results:\n['iNat_Idaho_Detection_Results.p', 'IDFG_Detection_Results_1.p', 'IDFG_Detection_Results_2.p', 'CCT_Detection_Results_2.p', 'CCT_Detection_Results_1.p']\n1. It seems to be, Detection_Results not include any test_images. Which is detection results for test_images?\n2. It seems to be, these are Bbox for full-size images, but I do not Know shapes for full-size images. So, How can I transfor the bbox to small-size images?\nThankyou
", "votes": null, "replies": [ { "id": 542421, "author_name": "sbeery", "author_url": "", "post_date": "06/03/2019 22:08:59", "content": "IDFGDetectionResults1 and 2 contain the results for the test images. The full image sizes are in the Annotation and Information files for each dataset.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "502416": "In the specs on Github, you mentioned that we can't use annotations for the test set. Does that mean that \n(a) we can't generate the bounding box data, and then load it in our model, or\n(b) we can't use a detection model such as FRCNN or MobileNet inside our kernel to generate bounding boxes?", "502510": "I mean that you cannot explicitly collect ground truth bounding boxes or species identifications for the test set. You are 100% allowed to run any detector you would like over the test set, as long as that detector is either the one we have provided (and we have released the detection results for that, so you should be able to use them directly without running inference) or a detector trained on only the data allowed for the competition (Imagenet, COCO, CCT, or iNaturalist 2017-2019)", "502524": ">(and we have released the detection results for that, so you should be able to use them directly without running inference)\n\nThank you for the quick answer! I was under the impression that the annotation results (bounding boxes) were only available for the data in the training sets, but not in the test sets. Are they both available on Github?", "502531": "Yep! They're all available on Github. But they probably aren't perfect! We don't have ground truth for the test data, so I don't have any metrics for the detection results.", "502533": "Sounds good! I probably missed it in this case. I'll take a closer look and try to make use of it. Thank you!", "502537": "NP :)", "509355": "Where you able to find the annotation results for the test sets? In the file provided on the GitHub (https://wildcamdrop.blob.core.windows.net/wildcamdropcontainer/Detection_Results.tar.gz) the test annotations file is empty. Is it available somewhere else?", "510107": "I will look into this, sounds like I uploaded the wrong file.", "510210": "I did. The ''Detection\\_Results'' zipped folder has now been updated to include the IDFG results, split across two files: 'IDFG\\_Detection\\_Results\\_1.p' and 'IDFG\\_Detection\\_Results\\_2.p'\n\nSo sorry for the confusion!", "510241": "Thank you!", "519043": "As for the classification, is the bounding box data helpful?", "542255": "sbeery \nI am sorry, I am confusing about the Detection_Results. \nThere are only 5 files in Detection_Results:\n['iNat_Idaho_Detection_Results.p', 'IDFG_Detection_Results_1.p', 'IDFG_Detection_Results_2.p', 'CCT_Detection_Results_2.p', 'CCT_Detection_Results_1.p']\n1. It seems to be, Detection_Results not include any test_images. Which is detection results for test_images?\n2. It seems to be, these are Bbox for full-size images, but I do not Know shapes for full-size images. So, How can I transfor the bbox to small-size images?\nThankyou", "542421": "IDFGDetectionResults1 and 2 contain the results for the test images. The full image sizes are in the Annotation and Information files for each dataset." }, "source": "meta" }