I believe the height values in the floor_info.json files for the floors on site 5d27075f03f801723c2e360f are incorrect - below are the adjustments I made to make the waypoints line up correctly with the image (these are approximate - it's difficult to tell exactly what they should be):
\nThis is the only site I noticed this for (for sites that have test paths - I haven't checked the other sites)
\nSite 5d27075f03f801723c2e360f
\nFloor B1
\n5d27075f03f801723c2e3618
\n252.25002019023339 => 240.0
Floor F1
\n5d27075f03f801723c2e3611
\n221.9711303099799 => 211
Floor F2
\n5d27075f03f801723c2e3614
\n222.0824509676249 => 211
Floor F3
\n5d27075f03f801723c2e3610
\n221.85980965595348 => 211
Floor F4
\n5d27075f03f801723c2e3612
\n221.85980965595348 => 211.0
Floor F5
\n5d27075f03f801723c2e3613
\n142.6008283613404 => 135.3
Floor F6
\n5d27075f03f801723c2e3615
\n142.48950770731392 => 135.3
Floor F7
\n5d27075f03f801723c2e3619
\n142.48950770731392 => 135.3
I believe the height values in the floor_info.json files for the floors on site 5d27075f03f801723c2e360f are incorrect - below are the adjustments I made to make the waypoints line up correctly with the image (these are approximate - it's difficult to tell exactly what they should be):
\nThis is the only site I noticed this for (for sites that have test paths - I haven't checked the other sites)
\nSite 5d27075f03f801723c2e360f
\nFloor B1
\n5d27075f03f801723c2e3618
\n252.25002019023339 => 240.0
Floor F1
\n5d27075f03f801723c2e3611
\n221.9711303099799 => 211
Floor F2
\n5d27075f03f801723c2e3614
\n222.0824509676249 => 211
Floor F3
\n5d27075f03f801723c2e3610
\n221.85980965595348 => 211
Floor F4
\n5d27075f03f801723c2e3612
\n221.85980965595348 => 211.0
Floor F5
\n5d27075f03f801723c2e3613
\n142.6008283613404 => 135.3
Floor F6
\n5d27075f03f801723c2e3615
\n142.48950770731392 => 135.3
Floor F7
\n5d27075f03f801723c2e3619
\n142.48950770731392 => 135.3
Oh, and it's also possible that it's the image that is actually not correct, but the image didn't look distorted to me; but I'm not 100% sure :)
", "rawMarkdown": "Oh, and it's also possible that it's the image that is actually not correct, but the image didn't look distorted to me; but I'm not 100% sure :)", "votes": null }, { "id": "1262596", "postDate": "04/04/2021 13:42:12", "content": "Look at my post, is related https://www.kaggle.com/c/indoor-location-navigation/discussion/230379
", "rawMarkdown": "Look at my post, is related https://www.kaggle.com/c/indoor-location-navigation/discussion/230379", "votes": null }, { "id": "1262601", "postDate": "04/04/2021 13:54:35", "content": "Ah, interesting - yep! it looks like that could be a similar issue (I haven't looked at that site yet)
", "rawMarkdown": "Ah, interesting - yep! it looks like that could be a similar issue (I haven't looked at that site yet)", "votes": null }, { "id": "1268590", "postDate": "04/09/2021 15:07:35", "content": "I wonder what will happen if you use snap to grid method with your scaling values. The LB score get better or worse?
", "rawMarkdown": "I wonder what will happen if you use snap to grid method with your scaling values. The LB score get better or worse?", "votes": null }, { "id": "1268666", "postDate": "04/09/2021 16:13:14", "content": "I don't think it would affect it because (I think) this only matters if you're converting between image pixels and meters. (so if you aren't using the images, then it shouldn't matter).
\nI believe the waypoint x,y values (used for snap to grid) are independent of this (could be wrong though!)
", "rawMarkdown": "I don't think it would affect it because (I think) this only matters if you're converting between image pixels and meters. (so if you aren't using the images, then it shouldn't matter). \n\nI believe the waypoint x,y values (used for snap to grid) are independent of this (could be wrong though!)", "votes": null }, { "id": "1268950", "postDate": "04/10/2021 01:27:40", "content": "I agree with you. The test labels would be consistent with train waypoints.
\nDo you find any specific rules (e.g., ratio) of the distortion? Are true height values always smaller?
", "rawMarkdown": "I agree with you. The test labels would be consistent with train waypoints. \n\nDo you find any specific rules (e.g., ratio) of the distortion? Are true height values always smaller?", "votes": null }, { "id": "1268960", "postDate": "04/10/2021 01:57:47", "content": "I was visualizing the waypoints on the map and noticed a shift (on that map only). I just played with some values (above) until the waypoints looked like they lined up on the map.
", "rawMarkdown": "I was visualizing the waypoints on the map and noticed a shift (on that map only). I just played with some values (above) until the waypoints looked like they lined up on the map.", "votes": null }, { "id": "1268970", "postDate": "04/10/2021 02:26:14", "content": "I see. Thanks!
", "rawMarkdown": "I see. Thanks!", "votes": null }, { "id": "1269148", "postDate": "04/10/2021 08:10:46", "content": "did you find any generic rule to move the waypoint to the correct position on image?
", "rawMarkdown": "did you find any generic rule to move the waypoint to the correct position on image?", "votes": null }, { "id": "1269450", "postDate": "04/10/2021 14:41:04", "content": "No, I tried a few things thinking I should be able to find exactly what the value should be, but couldn't make it match up - the values above are what I'm using now
", "rawMarkdown": "No, I tried a few things thinking I should be able to find exactly what the value should be, but couldn't make it match up - the values above are what I'm using now", "votes": null }, { "id": "1275131", "postDate": "04/16/2021 01:18:02", "content": "Chris, thanks for sharing your insights and adjustments to leverage the geojson file. I'm also working on this problem, because some grids are inside the shops, which looks strange to me. BTW, I sent a message to you using the contact system in kaggle! I would be happy if you could check the e-mail. 🙏
", "rawMarkdown": "Chris, thanks for sharing your insights and adjustments to leverage the geojson file. I'm also working on this problem, because some grids are inside the shops, which looks strange to me. BTW, I sent a message to you using the contact system in kaggle! I would be happy if you could check the e-mail. 🙏", "votes": null }, { "id": "1275153", "postDate": "04/16/2021 02:21:07", "content": "Yes, some grids appear inside shops but are supposed to be in the hallways (like the ones I mentioned above), though there are some that I believe are correctly inside the blue areas (like training path 5e15b007f4c3420006d5229f) - but I think those are rare, and most paths should never be inside the blue shops.
\n[also, I responded to your email - let me know if that didn't go through!]
", "rawMarkdown": "Yes, some grids appear inside shops but are supposed to be in the hallways (like the ones I mentioned above), though there are some that I believe are correctly inside the blue areas (like training path 5e15b007f4c3420006d5229f) - but I think those are rare, and most paths should never be inside the blue shops.\n\n[also, I responded to your email - let me know if that didn't go through!]", "votes": null }, { "id": "1275248", "postDate": "04/16/2021 05:58:52", "content": "Yeah, I guess it would be one of the key in this competition.
\nThanks for the quick reply, I responded to the e-mail :)
Thank you @chris62 for the insights. Have you found out whether it is actually the image or the height/width that is wrong?
", "rawMarkdown": "Thank you @chris62 for the insights. Have you found out whether it is actually the image or the height/width that is wrong?", "votes": null }, { "id": "1307892", "postDate": "05/14/2021 18:25:56", "content": "I'm still not 100% sure where the issue actually is, but for the post processing I'm doing I think it doesn't matter - those adjusted site values seem to work. (because I'm matching to the floorplan image, so it doesn't matter if I scale the site or the image for that).
\nIf you're using the site height/width values for something other than matching to the floorplan though (like the geo values), then I'm not sure, sorry!
", "rawMarkdown": "I'm still not 100% sure where the issue actually is, but for the post processing I'm doing I think it doesn't matter - those adjusted site values seem to work. (because I'm matching to the floorplan image, so it doesn't matter if I scale the site or the image for that).\n\nIf you're using the site height/width values for something other than matching to the floorplan though (like the geo values), then I'm not sure, sorry!", "votes": null }, { "id": "1307896", "postDate": "05/14/2021 18:36:20", "content": "Thank you for your reply :) I'll try to verify it somehow
", "rawMarkdown": "Thank you for your reply :) I'll try to verify it somehow", "votes": null }, { "id": "1307988", "postDate": "05/14/2021 20:52:28", "content": "It seems that one polygon's unit corresponds to 0.863 m (I have checked 4 floor-building pairs). This value is consistent with 5d27075f03f801723c2e360f's polygons and heights/widths.
", "rawMarkdown": "It seems that one polygon's unit corresponds to 0.863 m (I have checked 4 floor-building pairs). This value is consistent with 5d27075f03f801723c2e360f's polygons and heights/widths.", "votes": null } ], "comments": [ { "id": 1262584, "author_name": "chris62", "author_url": "", "post_date": "04/04/2021 13:24:59", "content": "Oh, and it's also possible that it's the image that is actually not correct, but the image didn't look distorted to me; but I'm not 100% sure :)
", "votes": null, "replies": [ { "id": 1307883, "author_name": "olaf2000", "author_url": "", "post_date": "05/14/2021 18:18:45", "content": "Thank you @chris62 for the insights. Have you found out whether it is actually the image or the height/width that is wrong?
", "votes": null, "replies": [] }, { "id": 1307892, "author_name": "chris62", "author_url": "", "post_date": "05/14/2021 18:25:56", "content": "I'm still not 100% sure where the issue actually is, but for the post processing I'm doing I think it doesn't matter - those adjusted site values seem to work. (because I'm matching to the floorplan image, so it doesn't matter if I scale the site or the image for that).
\nIf you're using the site height/width values for something other than matching to the floorplan though (like the geo values), then I'm not sure, sorry!
", "votes": null, "replies": [] }, { "id": 1307896, "author_name": "olaf2000", "author_url": "", "post_date": "05/14/2021 18:36:20", "content": "Thank you for your reply :) I'll try to verify it somehow
", "votes": null, "replies": [] }, { "id": 1307988, "author_name": "olaf2000", "author_url": "", "post_date": "05/14/2021 20:52:28", "content": "It seems that one polygon's unit corresponds to 0.863 m (I have checked 4 floor-building pairs). This value is consistent with 5d27075f03f801723c2e360f's polygons and heights/widths.
", "votes": null, "replies": [] } ] }, { "id": 1262596, "author_name": "enric1296", "author_url": "", "post_date": "04/04/2021 13:42:12", "content": "Look at my post, is related https://www.kaggle.com/c/indoor-location-navigation/discussion/230379
", "votes": null, "replies": [ { "id": 1262601, "author_name": "chris62", "author_url": "", "post_date": "04/04/2021 13:54:35", "content": "Ah, interesting - yep! it looks like that could be a similar issue (I haven't looked at that site yet)
", "votes": null, "replies": [] } ] }, { "id": 1268590, "author_name": "tomooinubushi", "author_url": "", "post_date": "04/09/2021 15:07:35", "content": "I wonder what will happen if you use snap to grid method with your scaling values. The LB score get better or worse?
", "votes": null, "replies": [ { "id": 1268666, "author_name": "chris62", "author_url": "", "post_date": "04/09/2021 16:13:14", "content": "I don't think it would affect it because (I think) this only matters if you're converting between image pixels and meters. (so if you aren't using the images, then it shouldn't matter).
\nI believe the waypoint x,y values (used for snap to grid) are independent of this (could be wrong though!)
", "votes": null, "replies": [] }, { "id": 1268950, "author_name": "tomooinubushi", "author_url": "", "post_date": "04/10/2021 01:27:40", "content": "I agree with you. The test labels would be consistent with train waypoints.
\nDo you find any specific rules (e.g., ratio) of the distortion? Are true height values always smaller?
", "votes": null, "replies": [] }, { "id": 1268960, "author_name": "chris62", "author_url": "", "post_date": "04/10/2021 01:57:47", "content": "I was visualizing the waypoints on the map and noticed a shift (on that map only). I just played with some values (above) until the waypoints looked like they lined up on the map.
", "votes": null, "replies": [] }, { "id": 1268970, "author_name": "tomooinubushi", "author_url": "", "post_date": "04/10/2021 02:26:14", "content": "I see. Thanks!
", "votes": null, "replies": [] }, { "id": 1269148, "author_name": "enric1296", "author_url": "", "post_date": "04/10/2021 08:10:46", "content": "did you find any generic rule to move the waypoint to the correct position on image?
", "votes": null, "replies": [] }, { "id": 1269450, "author_name": "chris62", "author_url": "", "post_date": "04/10/2021 14:41:04", "content": "No, I tried a few things thinking I should be able to find exactly what the value should be, but couldn't make it match up - the values above are what I'm using now
", "votes": null, "replies": [] } ] }, { "id": 1275131, "author_name": "mamasinkgs", "author_url": "", "post_date": "04/16/2021 01:18:02", "content": "Chris, thanks for sharing your insights and adjustments to leverage the geojson file. I'm also working on this problem, because some grids are inside the shops, which looks strange to me. BTW, I sent a message to you using the contact system in kaggle! I would be happy if you could check the e-mail. 🙏
", "votes": null, "replies": [ { "id": 1275153, "author_name": "chris62", "author_url": "", "post_date": "04/16/2021 02:21:07", "content": "Yes, some grids appear inside shops but are supposed to be in the hallways (like the ones I mentioned above), though there are some that I believe are correctly inside the blue areas (like training path 5e15b007f4c3420006d5229f) - but I think those are rare, and most paths should never be inside the blue shops.
\n[also, I responded to your email - let me know if that didn't go through!]
", "votes": null, "replies": [] }, { "id": 1275248, "author_name": "mamasinkgs", "author_url": "", "post_date": "04/16/2021 05:58:52", "content": "Yeah, I guess it would be one of the key in this competition.
\nThanks for the quick reply, I responded to the e-mail :)