From sukhad.anand at gmail.com Wed Dec 21 03:12:35 2016 From: sukhad.anand at gmail.com (Sukhad Anand) Date: Wed, 21 Dec 2016 13:42:35 +0530 Subject: [Kwiver-users] Important Message-ID: when i run track_features on a public dataset , after each frame it outputs writing invalid homography. And when i run adjust_track then it displays this : no landmark projections for new camera current reprojection RMSE: -nan frame 0 - num landmarks = 0 frame 0 uses reference 0 and further processing does not takes place as if the program goes into infinite loop. Please suggest how to get rid of such problems. -------------- next part -------------- An HTML attachment was scrubbed... URL: From matt.leotta at kitware.com Wed Dec 21 09:00:13 2016 From: matt.leotta at kitware.com (Matt Leotta) Date: Wed, 21 Dec 2016 09:00:13 -0500 Subject: [Kwiver-users] Important In-Reply-To: References: Message-ID: Sukhad, My guess is that the maptk_track_features tool is not finding enough features or enough feature matches. Which version of MAP-Tk are you using? What is the public dataset? --Matt On Wed, Dec 21, 2016 at 3:12 AM, Sukhad Anand wrote: > when i run track_features on a public dataset , after each frame it > outputs writing invalid homography. > And when i run adjust_track then it displays this : > > no landmark projections for new camera > current reprojection RMSE: -nan > frame 0 - num landmarks = 0 > frame 0 uses reference 0 > and further processing does not takes place as if the program goes into > infinite loop. > > Please suggest how to get rid of such problems. > > _______________________________________________ > Kwiver-users mailing list > Kwiver-users at public.kitware.com > http://public.kitware.com/mailman/listinfo/kwiver-users > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From matt.leotta at kitware.com Wed Dec 21 12:11:35 2016 From: matt.leotta at kitware.com (Matthew Leotta) Date: Wed, 21 Dec 2016 12:11:35 -0500 Subject: [Kwiver-users] Important In-Reply-To: References: Message-ID: > On Dec 21, 2016, at 10:42 AM, Sukhad Anand wrote: > > Matt, > > I am using MAP-Tk 0.8.0. I suggest you use the latest MAP-Tk, 0.9.0. It has options for much improved loop closure that you will probably need. > > This is the link for the dataset which i am using. > https://www.sensefly.com/drones/example-datasets.html Thanks for the link. Which specific data set are you trying? It would be helpful to know in order to reproduce this. > > Is it possible to run this tool on a drone making linear motion instead of orbital motion as in kwiwer_fmv_set1 ? Yes, but we?ve done far less testing on that type of data. Also the current pipeline is optimized for video instead of images capture at a lower frame rate. You will likely need to adjust the configuration settings to make it work in these cases. > > i would be obliged if you could provide the link for detailed explanation of the loop closures used by MAP-Tk There a multiple algorithms that you can configure for different needs, and the whole framework is extensible so you can even write your own loop closure algorithm to plug in. The current ?best? available algorithm (as of 0.9,0) is a key frame based method. It?s an image-to-image matching approach. In short, the algorithm matches each frame to the previous frames and a list of key frames. When none of the key frames match it establishes a new key frame. it?s a bit more complicated, but that?s the basic idea. If I can find some time I?ll try get things running on these sensefly data sets. ?Matt > > Thanks > > On Wed, Dec 21, 2016 at 7:30 PM, Matt Leotta > wrote: > Sukhad, > > My guess is that the maptk_track_features tool is not finding enough features or enough feature matches. Which version of MAP-Tk are you using? What is the public dataset? > > --Matt > > On Wed, Dec 21, 2016 at 3:12 AM, Sukhad Anand > wrote: > when i run track_features on a public dataset , after each frame it outputs writing invalid homography. > And when i run adjust_track then it displays this : > > no landmark projections for new camera > current reprojection RMSE: -nan > frame 0 - num landmarks = 0 > frame 0 uses reference 0 > and further processing does not takes place as if the program goes into infinite loop. > > Please suggest how to get rid of such problems. > > _______________________________________________ > Kwiver-users mailing list > Kwiver-users at public.kitware.com > http://public.kitware.com/mailman/listinfo/kwiver-users > > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From matt.leotta at kitware.com Wed Dec 21 17:15:41 2016 From: matt.leotta at kitware.com (Matt Leotta) Date: Wed, 21 Dec 2016 17:15:41 -0500 Subject: [Kwiver-users] Important In-Reply-To: References: Message-ID: Please reply to the kwiver-users address as well as mine. Someone else on the list may chime in with additional info. My answers are inline below. On Wed, Dec 21, 2016 at 4:11 PM, Sukhad Anand wrote: > I am using the first dataset in the list "Horizontal mappong with eBee". > I'll try to process that and see what issues I have. > > > Is it possible to obtain the trajectory of the camera on the image itself? > Yes it is, up to a similarity transform. In other words, you won't know the final scale, geo-position, or geo-orientation of the solution without either GPS or ground control points, but you can get a relative trajectory of the cameras. > > For what purpose is .pos file given as input in bundle_adjust_tracks? > The .pos files are currently the format for input metadata (GPS, INS) that we can use for initialization and geo-location. These are optional inputs. > > Is it possible to use the tool for real time navigation? > No, not yet. Our current focus is offline optimization for best quality 3D reconstruction. We are interesting in exploring real-time in the future, but don't have any funding to do that at the moment. > > How can i produce geo coordinates using current camera view? > You'll need either GPS measurements for some of the images, or you'll need to manually select some ground control points. This will allow you to get a geo-registered result. Unfortunately, that part is not very intuitive or well documented at the moment, but we are working on improving it. > > How to load my own plugins? > The plugin framework is still evolving (this is a relatively new and quickly growing project). In MAP-Tk 0.8.0 and 0.9.0 the plugins are installed in ${INSTALL_PATH}/lib/maptk. Everything in that directory is loaded. You can add plugins there or specify additional paths with the environment variable KWIVER_PLUGIN_PATH. As of the master branch of MAP-Tk (and future releases) all the MAP-Tk plugins have moved into the KWIVER project. The install path might change before the next release, but the environment variable is still the same. The next obvious question is "How do I create my own plugin?". We'll need to write a blog post on that in the future and put together a template to get people started, but that hasn't happened yet. For now, the best way is to look at the existing plugins. These plugins have been called "arrows" after moving into KWIVER. > > It would be a great help if you could provide a good tutorial link for the > tool. > Here is one tutorial: https://blog.kitware.com/getting-started-with-map-tk-for-aerial-photogrammetry/ and this one describes more recent changes: https://blog.kitware.com/map-tk-0-9-0-release-provides-dense-3d-better-loop-closure-and-opencv-3-support/ We'll try to release more tutorials and updates like these in the future as things continue to change. --Matt > > > On Wed, Dec 21, 2016 at 10:41 PM, Matthew Leotta > wrote: > >> >> On Dec 21, 2016, at 10:42 AM, Sukhad Anand >> wrote: >> >> Matt, >> >> I am using MAP-Tk 0.8.0. >> >> >> I suggest you use the latest MAP-Tk, 0.9.0. It has options for much >> improved loop closure that you will probably need. >> >> >> This is the link for the dataset which i am using. >> https://www.sensefly.com/drones/example-datasets.html >> >> >> Thanks for the link. Which specific data set are you trying? It would >> be helpful to know in order to reproduce this. >> >> >> Is it possible to run this tool on a drone making linear motion instead >> of orbital motion as in kwiwer_fmv_set1 ? >> >> >> Yes, but we?ve done far less testing on that type of data. Also the >> current pipeline is optimized for video instead of images capture at a >> lower frame rate. You will likely need to adjust the configuration >> settings to make it work in these cases. >> >> >> i would be obliged if you could provide the link for detailed explanation >> of the loop closures used by MAP-Tk >> >> >> There a multiple algorithms that you can configure for different needs, >> and the whole framework is extensible so you can even write your own loop >> closure algorithm to plug in. The current ?best? available algorithm (as >> of 0.9,0) is a key frame based method. It?s an image-to-image matching >> approach. In short, the algorithm matches each frame to the previous >> frames and a list of key frames. When none of the key frames match it >> establishes a new key frame. it?s a bit more complicated, but that?s the >> basic idea. >> >> If I can find some time I?ll try get things running on these sensefly >> data sets. >> >> ?Matt >> >> >> Thanks >> >> On Wed, Dec 21, 2016 at 7:30 PM, Matt Leotta >> wrote: >> >>> Sukhad, >>> >>> My guess is that the maptk_track_features tool is not finding enough >>> features or enough feature matches. Which version of MAP-Tk are you >>> using? What is the public dataset? >>> >>> --Matt >>> >>> On Wed, Dec 21, 2016 at 3:12 AM, Sukhad Anand >>> wrote: >>> >>>> when i run track_features on a public dataset , after each frame it >>>> outputs writing invalid homography. >>>> And when i run adjust_track then it displays this : >>>> >>>> no landmark projections for new camera >>>> current reprojection RMSE: -nan >>>> frame 0 - num landmarks = 0 >>>> frame 0 uses reference 0 >>>> and further processing does not takes place as if the program goes >>>> into infinite loop. >>>> >>>> Please suggest how to get rid of such problems. >>>> >>>> _______________________________________________ >>>> Kwiver-users mailing list >>>> Kwiver-users at public.kitware.com >>>> http://public.kitware.com/mailman/listinfo/kwiver-users >>>> >>>> >>> >> >> > -------------- next part -------------- An HTML attachment was scrubbed... URL: From sukhad.anand at gmail.com Thu Dec 22 13:14:57 2016 From: sukhad.anand at gmail.com (Sukhad Anand) Date: Thu, 22 Dec 2016 23:44:57 +0530 Subject: [Kwiver-users] Important In-Reply-To: References: Message-ID: Can you please provide an executable file or suggest the steps which i should follow and files to use to obtain trajectory of the camera on the image itself as i am a newbie? What does the values in .krtd files signify? Can u please provide an executable file or suggest steps to follow to convert the information obtained from vision into gps coordinates? On Thu, Dec 22, 2016 at 3:45 AM, Matt Leotta wrote: > Please reply to the kwiver-users address as well as mine. Someone else on > the list may chime in with additional info. My answers are inline below. > > > On Wed, Dec 21, 2016 at 4:11 PM, Sukhad Anand > wrote: > >> I am using the first dataset in the list "Horizontal mappong with eBee". >> > > I'll try to process that and see what issues I have. > > >> >> >> Is it possible to obtain the trajectory of the camera on the image itself? >> > > Yes it is, up to a similarity transform. In other words, you won't know > the final scale, geo-position, or geo-orientation of the solution without > either GPS or ground control points, but you can get a relative trajectory > of the cameras. > > >> >> For what purpose is .pos file given as input in bundle_adjust_tracks? >> > > The .pos files are currently the format for input metadata (GPS, INS) that > we can use for initialization and geo-location. These are optional inputs. > > >> >> Is it possible to use the tool for real time navigation? >> > > No, not yet. Our current focus is offline optimization for best quality > 3D reconstruction. We are interesting in exploring real-time in the > future, but don't have any funding to do that at the moment. > > >> >> How can i produce geo coordinates using current camera view? >> > > You'll need either GPS measurements for some of the images, or you'll need > to manually select some ground control points. This will allow you to get > a geo-registered result. Unfortunately, that part is not very intuitive or > well documented at the moment, but we are working on improving it. > > >> >> How to load my own plugins? >> > > The plugin framework is still evolving (this is a relatively new and > quickly growing project). In MAP-Tk 0.8.0 and 0.9.0 the plugins are > installed in ${INSTALL_PATH}/lib/maptk. Everything in that directory is > loaded. You can add plugins there or specify additional paths with the > environment variable KWIVER_PLUGIN_PATH. As of the master branch of MAP-Tk > (and future releases) all the MAP-Tk plugins have moved into the KWIVER > project. The install path might change before the next release, but the > environment variable is still the same. > > The next obvious question is "How do I create my own plugin?". We'll need > to write a blog post on that in the future and put together a template to > get people started, but that hasn't happened yet. For now, the best way is > to look at the existing plugins. These plugins have been called "arrows" > after moving into KWIVER. > > >> >> It would be a great help if you could provide a good tutorial link for >> the tool. >> > > Here is one tutorial: > > https://blog.kitware.com/getting-started-with-map-tk- > for-aerial-photogrammetry/ > > and this one describes more recent changes: > > https://blog.kitware.com/map-tk-0-9-0-release-provides- > dense-3d-better-loop-closure-and-opencv-3-support/ > > We'll try to release more tutorials and updates like these in the future > as things continue to change. > > --Matt > > > >> >> >> On Wed, Dec 21, 2016 at 10:41 PM, Matthew Leotta > > wrote: >> >>> >>> On Dec 21, 2016, at 10:42 AM, Sukhad Anand >>> wrote: >>> >>> Matt, >>> >>> I am using MAP-Tk 0.8.0. >>> >>> >>> I suggest you use the latest MAP-Tk, 0.9.0. It has options for much >>> improved loop closure that you will probably need. >>> >>> >>> This is the link for the dataset which i am using. >>> https://www.sensefly.com/drones/example-datasets.html >>> >>> >>> Thanks for the link. Which specific data set are you trying? It would >>> be helpful to know in order to reproduce this. >>> >>> >>> Is it possible to run this tool on a drone making linear motion instead >>> of orbital motion as in kwiwer_fmv_set1 ? >>> >>> >>> Yes, but we?ve done far less testing on that type of data. Also the >>> current pipeline is optimized for video instead of images capture at a >>> lower frame rate. You will likely need to adjust the configuration >>> settings to make it work in these cases. >>> >>> >>> i would be obliged if you could provide the link for detailed >>> explanation of the loop closures used by MAP-Tk >>> >>> >>> There a multiple algorithms that you can configure for different needs, >>> and the whole framework is extensible so you can even write your own loop >>> closure algorithm to plug in. The current ?best? available algorithm (as >>> of 0.9,0) is a key frame based method. It?s an image-to-image matching >>> approach. In short, the algorithm matches each frame to the previous >>> frames and a list of key frames. When none of the key frames match it >>> establishes a new key frame. it?s a bit more complicated, but that?s the >>> basic idea. >>> >>> If I can find some time I?ll try get things running on these sensefly >>> data sets. >>> >>> ?Matt >>> >>> >>> Thanks >>> >>> On Wed, Dec 21, 2016 at 7:30 PM, Matt Leotta >>> wrote: >>> >>>> Sukhad, >>>> >>>> My guess is that the maptk_track_features tool is not finding enough >>>> features or enough feature matches. Which version of MAP-Tk are you >>>> using? What is the public dataset? >>>> >>>> --Matt >>>> >>>> On Wed, Dec 21, 2016 at 3:12 AM, Sukhad Anand >>>> wrote: >>>> >>>>> when i run track_features on a public dataset , after each frame it >>>>> outputs writing invalid homography. >>>>> And when i run adjust_track then it displays this : >>>>> >>>>> no landmark projections for new camera >>>>> current reprojection RMSE: -nan >>>>> frame 0 - num landmarks = 0 >>>>> frame 0 uses reference 0 >>>>> and further processing does not takes place as if the program goes >>>>> into infinite loop. >>>>> >>>>> Please suggest how to get rid of such problems. >>>>> >>>>> _______________________________________________ >>>>> Kwiver-users mailing list >>>>> Kwiver-users at public.kitware.com >>>>> http://public.kitware.com/mailman/listinfo/kwiver-users >>>>> >>>>> >>>> >>> >>> >> > -------------- next part -------------- An HTML attachment was scrubbed... URL: From sukhad.anand at gmail.com Fri Dec 23 03:33:25 2016 From: sukhad.anand at gmail.com (Sukhad Anand) Date: Fri, 23 Dec 2016 14:03:25 +0530 Subject: [Kwiver-users] errors on a dataset Message-ID: I ran bundle_adjust_tracks on my own dataset The processing hangs after displaying the below output E matrix num inliers = 12257/12257 median scale = 1 no landmark projections for new camera current reprojection RMSE: -nan frame 0 - num landmarks = 0 frame 0 uses reference 0 please suggest how to correct this -------------- next part -------------- An HTML attachment was scrubbed... URL: From sukhad.anand at gmail.com Fri Dec 23 03:46:55 2016 From: sukhad.anand at gmail.com (Sukhad Anand) Date: Fri, 23 Dec 2016 14:16:55 +0530 Subject: [Kwiver-users] (no subject) Message-ID: what does the camera sample rate signify in the bundle_adjust_tracks.conf ? -------------- next part -------------- An HTML attachment was scrubbed... URL: