Hello，
\nI noticed that the training set includes 7 scenes (church, dioscuri, lizard, multi-temporal-temple-baalshamin, pond, transp_obj_glass_cup, and transp_obj_glass_cylinder), and the available sample test set also contains church and lizard.
\nMy question is, are the scenes in the test set exactly the same as those in the training set? If not, is the set of training scenes a subset of the test scenes?
\nFurthermore, for instance, with the scene named \"church\", can we perform joint optimization/reconstruction on the church from the training set with the church from the test set?
", "messages": [ { "id": 2743342, "postDate": "2024-04-09T11:50:56.877Z", "content": "Hello，
\nI noticed that the training set includes 7 scenes (church, dioscuri, lizard, multi-temporal-temple-baalshamin, pond, transp_obj_glass_cup, and transp_obj_glass_cylinder), and the available sample test set also contains church and lizard.
\nMy question is, are the scenes in the test set exactly the same as those in the training set? If not, is the set of training scenes a subset of the test scenes?
\nFurthermore, for instance, with the scene named \"church\", can we perform joint optimization/reconstruction on the church from the training set with the church from the test set?
", "rawMarkdown": "Hello，\n\nI noticed that the training set includes 7 scenes (church, dioscuri, lizard, multi-temporal-temple-baalshamin, pond, transp_obj_glass_cup, and transp_obj_glass_cylinder), and the available sample test set also contains church and lizard.\n\nMy question is, are the scenes in the test set exactly the same as those in the training set? If not, is the set of training scenes a subset of the test scenes?\n\nFurthermore, for instance, with the scene named \"church\", can we perform joint optimization/reconstruction on the church from the training set with the church from the test set?", "votes": 4 }, { "id": 2743721, "postDate": "2024-04-09T15:31:09.243Z", "content": "There is no overlap between training and test set.
\nThe \"sample test set\" is there just for helping participants to debug, nothing more.
@oldufo Just to clarify, do you mean that there aren't overlapping images (which is to be expected) or that there aren't overlapping scenes between train and test? e.g. is there no \"church\" scene in the test (with different images to those of the \"church\" scene in train of course)?
", "rawMarkdown": "@oldufo Just to clarify, do you mean that there aren't overlapping images (which is to be expected) or that there aren't overlapping scenes between train and test? e.g. is there no \"church\" scene in the test (with different images to those of the \"church\" scene in train of course)? ", "replies": [ { "id": 2745097, "postDate": "2024-04-10T11:24:51.233Z", "content": "There are no overlapping scenes, of course.
\nThere is no \"church\" scene in the test, for example
Understood, thanks for clarifying!
", "rawMarkdown": "Understood, thanks for clarifying!" }, { "id": 2745281, "postDate": "2024-04-10T14:18:43.130Z", "content": "@oldufo , So you mean the training set is provided to find best set parameters for SfM pipeline using various metadata and We should perform complete pipeline on test set (based on meta data and our chosen set of parameters on different scene entirely) and produce submission.csv? Please correct me if i am wrong. I just wanted to have a bit more clarity.
", "rawMarkdown": "@oldufo , So you mean the training set is provided to find best set parameters for SfM pipeline using various metadata and We should perform complete pipeline on test set (based on meta data and our chosen set of parameters on different scene entirely) and produce submission.csv? Please correct me if i am wrong. I just wanted to have a bit more clarity. ", "isDeleted": true }, { "id": 2745311, "postDate": "2024-04-10T14:36:52.443Z", "content": "@sapainca I am not sure that I understand your question. The training data is provided to give participants an example of which problems they can expect in the test set. For example, repeated patterns, day-night, cross-seasons, transparent objects, nature scenes, etc.
\nYou also may use it for training or tuning hyperparameters.
\nAnd in the test set you run your SfM pipeline on a completely different (and hidden) set of scenes. Does it become more clear now?
Yeah. Very clear now. Thanks for prompt response @oldufo
", "rawMarkdown": "Yeah. Very clear now. Thanks for prompt response @oldufo ", "isDeleted": true }, { "id": 2745470, "postDate": "2024-04-10T16:40:03.440Z", "content": "Very helpful~Thanks
", "rawMarkdown": "Very helpful~Thanks" }, { "id": 2746979, "postDate": "2024-04-11T15:23:44.420Z", "content": "@oldufo One further query, Is the file categories csv file (given in train folder) also provided with public/private test set folder?
", "rawMarkdown": "@oldufo One further query, Is the file categories csv file (given in train folder) also provided with public/private test set folder? ", "votes": 1, "isDeleted": true }, { "id": 2747106, "postDate": "2024-04-11T17:22:37.923Z", "content": "I believe that is a question to @sohier, as I am not sure.
", "rawMarkdown": "I believe that is a question to @sohier, as I am not sure. ", "votes": 1 }, { "id": 2747353, "postDate": "2024-04-11T20:14:41.300Z", "content": "Also had that question, assumed that since there is a categories.csv in the test dir, it would be updated with the hidden test. Would appreciate a confirmation
Do you have an answer to the question of whether categories.csv exists? Or can you ask him this question for us？If you can reply, thank you very much
", "rawMarkdown": "Do you have an answer to the question of whether categories.csv exists? Or can you ask him this question for us？If you can reply, thank you very much" } ] } ] }, { "id": 2745466, "postDate": "2024-04-10T16:39:12.630Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 2743721, "author_name": "old-ufo", "author_url": "", "post_date": "2024-04-09T15:31:09.243000", "content": "There is no overlap between training and test set.
\nThe \"sample test set\" is there just for helping participants to debug, nothing more.
@oldufo Just to clarify, do you mean that there aren't overlapping images (which is to be expected) or that there aren't overlapping scenes between train and test? e.g. is there no \"church\" scene in the test (with different images to those of the \"church\" scene in train of course)?
", "votes": 0, "replies": [ { "id": 2745097, "author_name": "old-ufo", "author_url": "", "post_date": "2024-04-10T11:24:51.233000", "content": "There are no overlapping scenes, of course.
\nThere is no \"church\" scene in the test, for example
Understood, thanks for clarifying!
", "votes": 0, "replies": [] }, { "id": 2745281, "author_name": "", "author_url": "", "post_date": "2024-04-10T14:18:43.130000", "content": "@oldufo , So you mean the training set is provided to find best set parameters for SfM pipeline using various metadata and We should perform complete pipeline on test set (based on meta data and our chosen set of parameters on different scene entirely) and produce submission.csv? Please correct me if i am wrong. I just wanted to have a bit more clarity.
", "votes": 0, "replies": [] }, { "id": 2745311, "author_name": "old-ufo", "author_url": "", "post_date": "2024-04-10T14:36:52.443000", "content": "@sapainca I am not sure that I understand your question. The training data is provided to give participants an example of which problems they can expect in the test set. For example, repeated patterns, day-night, cross-seasons, transparent objects, nature scenes, etc.
\nYou also may use it for training or tuning hyperparameters.
\nAnd in the test set you run your SfM pipeline on a completely different (and hidden) set of scenes. Does it become more clear now?
Yeah. Very clear now. Thanks for prompt response @oldufo
", "votes": 0, "replies": [] }, { "id": 2745470, "author_name": "sunny", "author_url": "", "post_date": "2024-04-10T16:40:03.440000", "content": "Very helpful~Thanks
", "votes": 0, "replies": [] }, { "id": 2746979, "author_name": "", "author_url": "", "post_date": "2024-04-11T15:23:44.420000", "content": "@oldufo One further query, Is the file categories csv file (given in train folder) also provided with public/private test set folder?
", "votes": 1, "replies": [] }, { "id": 2747106, "author_name": "old-ufo", "author_url": "", "post_date": "2024-04-11T17:22:37.923000", "content": "I believe that is a question to @sohier, as I am not sure.
", "votes": 1, "replies": [] }, { "id": 2747353, "author_name": "asarvazyan", "author_url": "", "post_date": "2024-04-11T20:14:41.300000", "content": "Also had that question, assumed that since there is a categories.csv in the test dir, it would be updated with the hidden test. Would appreciate a confirmation
Do you have an answer to the question of whether categories.csv exists? Or can you ask him this question for us？If you can reply, thank you very much
", "votes": 0, "replies": [] } ] } ] }, { "id": 2745466, "author_name": "", "author_url": "", "post_date": "2024-04-10T16:39:12.630000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2743342": "Hello，\n\nI noticed that the training set includes 7 scenes (church, dioscuri, lizard, multi-temporal-temple-baalshamin, pond, transp_obj_glass_cup, and transp_obj_glass_cylinder), and the available sample test set also contains church and lizard.\n\nMy question is, are the scenes in the test set exactly the same as those in the training set? If not, is the set of training scenes a subset of the test scenes?\n\nFurthermore, for instance, with the scene named \"church\", can we perform joint optimization/reconstruction on the church from the training set with the church from the test set?", "2743721": "There is no overlap between training and test set. \nThe \"sample test set\" is there just for helping participants to debug, nothing more.", "2745466": "" } }