Re: [Bf-committers] Blender and laser scanner - amendment
Hi, Indeed, adding color CD layer to vertices is not a great deal. For render, I think the simplest solution would be to tweak a bit particles (we already have billboards, just would have to make them be able to take the color from actual vertices and not only from loops), a system with as much particles as vertices, emitted from those all at once and with lifetime=scene duration is a suitable way to render a points cloud. With a decent modern computer, Blender can render several millions of such billboards with no problem. :) And we could also have some tool to convert points clouds to actual meshes + textures, which could generate much faster objects for rendering... So, especially if Jace already did part of the work (btw, Jace, any patch, branch or so available?), to get a basic usable and rederable solution should not be *that* hard! Best regards, Bastien On 30/04/2013 01:30, Isaac Lenton wrote: Hi I looked into this for a project I was working on for visualising point clouds generated from computational physics simulations. I recall that I managed to get the points rendering in Blender Render with a few quick hacks and duplicating the colour data onto individual vertices (maybe not the best idea). I ended up switching to VTK and ParaView for most of my scientific data visualisation, but I would be interested if someone were to implement a basic pipeline as described by Jace Priester but with a step 6: 6) Support for community add-ons (through python) for preforming basic data/point processing/rendering operations. ParaView provides a large selection of operators, some of which would be trivial to implement in blender if a pipeline was set-up. This idea might also be of interest to Shabbir Marzban, whose recent GSoC email to the mailing list included a proposal to integrate Bundler: 1) Automated rigid structure reconstruction from set of un-ordered images: A very common structure from motion (SFM) tool, Bundler[1] can be used to recover 3D point cloud from such images. Thanks, Isaac Lenton PS: The mailing list seems to be blocking some of my emails, not sure if this will get through. On Mon, 2013-04-29 at 15:02 -0700, Jace Priester wrote: I wrote in point cloud support for Blender over a year ago, but I have been told repeatedly by the developers that there is no interest in the community and therefore they aren't interested in implementing it. Hopefully another voice will help with that. The code isn't extensive as is. It requires a custom data layer to support true per-vertex color (as Blender currently implements vertex color as part of face data - not vertex data). There's a bit of extra code in the mesh drawing functions to access the colors in the custom data layer, and then it draws points to the screen using the set colors. As a bonus, it also supports per-vertex sizing which allows for point distance attenuation (which is necessary for a good appearance). My current code does *not* integrate with any rendering engine - it only works in the editor window - and I gave up trying to get help taking it any further, as I don't use it that often anyway. If a developer knowledgeable enough about BI/cycles wants to hop in and help with finishing this, I'd be happy to collaborate. I see no need for blender to be a point cloud editing program. Meshlab and others are better suited to that, as that is specifically what they are built for. But the ability to import and display point cloud data, in my opinion, is vital to Blender being usable in more fields than it is now. So that said, I think short term and easily attainable goals should be to 1) import point cloud data in a couple common formats (I have done this already) 2) display points with colors and distance attentuation (also done) 3) export point cloud data (also done) 4) some user interface controls for turning colors on/off, adjusting displayed vertex size and attenuation (have not done any of this) 5) support for render at least in BI On Mon, Apr 29, 2013 at 2:19 PM, Antoine Cottinacot...@ultiscan.comwrote: Hi y'all, Sorry for the repost this might be obvious or not, but after re-reading my previous post, I feel that need to mention that our goal is not to produce a tool dedicate to our sole use. Our goal is to develop a tool for us and the Blender community that will be release under GPL. Cheers. Antoine ### My name is Antoine Cottin, I'm a research scientist specialized in active remote sensing applied to earth sciences. For the last 8 years I've been working for/with a laser scanner manufacturer developing processing tools chain for laser scanner data processing. I have an extended knowledge in the various laser scanner systems (hardware and software) and their associated data format. I'm the founder of Ultiscan, a French base company, specialized in laser scanning services.
Re: [Bf-committers] Blender and laser scanner - amendment
Custom data layer was recommended by ideasman and it serves the purpose very well. I added a method for python to interact with the CDL and flash new data to it all in one shot instead of updating it one index at a time, which essentially made color import have no noticeable impact on import time. By using GLSL to draw the points in the editor I also don't notice any delay in speed rendering the editor. You guys are underestimating the difficulty of converting point clouds to textured meshes. This is an area which is still in research and development in many other programs and research projects. There are entire software libraries and programs devoted to that task and frankly all of them fail pretty horribly given anything except laboratory-grade clean point cloud data. Under normal circumstances with a normal amount of noise and holes in the data, point clouds do not mesh well. I would be all for adding it to Blender if there were mature code to plug in, but there is not, and it's a bad idea to merge in a solution that is going to produce poor results most of the time. Additionally, structure from motion is also far beyond the scope of Blender. SFM point clouds work much better than point cloud to mesh, but again there are entire software libraries and programs devoted to the task. Blender has absolutely no mechanisms as-is for handling any part of an SFM pipeline. As far as: 6) Support for community add-ons (through python) for preforming basic data/point processing/rendering operations. ..well, that's what bpy is there for. Vertex data is easily accessible as-is. I hate to be the one that shoots down all the ideas but what you're talking about is massive changes in the direction of the program. That isn't going to happen without backing from the core developers, which I assure you isn't there. Restricting this to import-export-render of point colors is a very attainable goal. On Tue, Apr 30, 2013 at 5:39 AM, Bastien Montagne montagn...@wanadoo.frwrote: Hi, Indeed, adding color CD layer to vertices is not a great deal. For render, I think the simplest solution would be to tweak a bit particles (we already have billboards, just would have to make them be able to take the color from actual vertices and not only from loops), a system with as much particles as vertices, emitted from those all at once and with lifetime=scene duration is a suitable way to render a points cloud. With a decent modern computer, Blender can render several millions of such billboards with no problem. :) And we could also have some tool to convert points clouds to actual meshes + textures, which could generate much faster objects for rendering... So, especially if Jace already did part of the work (btw, Jace, any patch, branch or so available?), to get a basic usable and rederable solution should not be *that* hard! Best regards, Bastien On 30/04/2013 01:30, Isaac Lenton wrote: Hi I looked into this for a project I was working on for visualising point clouds generated from computational physics simulations. I recall that I managed to get the points rendering in Blender Render with a few quick hacks and duplicating the colour data onto individual vertices (maybe not the best idea). I ended up switching to VTK and ParaView for most of my scientific data visualisation, but I would be interested if someone were to implement a basic pipeline as described by Jace Priester but with a step 6: 6) Support for community add-ons (through python) for preforming basic data/point processing/rendering operations. ParaView provides a large selection of operators, some of which would be trivial to implement in blender if a pipeline was set-up. This idea might also be of interest to Shabbir Marzban, whose recent GSoC email to the mailing list included a proposal to integrate Bundler: 1) Automated rigid structure reconstruction from set of un-ordered images: A very common structure from motion (SFM) tool, Bundler[1] can be used to recover 3D point cloud from such images. Thanks, Isaac Lenton PS: The mailing list seems to be blocking some of my emails, not sure if this will get through. On Mon, 2013-04-29 at 15:02 -0700, Jace Priester wrote: I wrote in point cloud support for Blender over a year ago, but I have been told repeatedly by the developers that there is no interest in the community and therefore they aren't interested in implementing it. Hopefully another voice will help with that. The code isn't extensive as is. It requires a custom data layer to support true per-vertex color (as Blender currently implements vertex color as part of face data - not vertex data). There's a bit of extra code in the mesh drawing functions to access the colors in the custom data layer, and then it draws points to the screen using the set colors. As a bonus, it also supports
Re: [Bf-committers] Blender and laser scanner - amendment
I agree that Blender is not a point cloud editor but adding point cloud visualization would be a nice add-on and need to stick to import-export-render. Concerning the SFM, I agree that they are dedicate softwares out there that do the job more or less nicely, however, SFM and camera tracking with 3D reconstruction use the same approach - photogrammetry. --- Ultiscan SAS 6 rue de Molsheim 67000 Strasbourg France www.ultiscan.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers
Re: [Bf-committers] Blender and laser scanner - amendment
Sort of. Camera tracking is based on following points through a video. As it is in blender that means manually marking points, automatic tracking, and manual adjustment to clean it up. SFM from photos is based on determining arbitrary and automatic point correspondence rather than 2D point tracking. I suppose that means that to implement SFM, Blender already has at least some of the needed code, but it's still missing point correspondence entirely. I'd skip all the slow methods and look at SiftGPU/SURF for quick point correspondence determination. As a bonus this would probably be a great help to the camera tracking portion as well. My last set of point cloud code was for some early 2.6 version and custom data layers have been added since then which breaks my code. I'll put together a patch for the current version and get back to you. On Tue, Apr 30, 2013 at 9:06 AM, Antoine Cottin acot...@ultiscan.comwrote: I agree that Blender is not a point cloud editor but adding point cloud visualization would be a nice add-on and need to stick to import-export-render. Concerning the SFM, I agree that they are dedicate softwares out there that do the job more or less nicely, however, SFM and camera tracking with 3D reconstruction use the same approach - photogrammetry. --- Ultiscan SAS 6 rue de Molsheim 67000 Strasbourg France www.ultiscan.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers -- -- Jace Priester Threespace Imaging jacepries...@threespaceimaging.com 559-284-0904 -- ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers
Re: [Bf-committers] Blender and laser scanner - amendment
Hi I looked into this for a project I was working on for visualising point clouds generated from computational physics simulations. I recall that I managed to get the points rendering in Blender Render with a few quick hacks and duplicating the colour data onto individual vertices (maybe not the best idea). I ended up switching to VTK and ParaView for most of my scientific data visualisation, but I would be interested if someone were to implement a basic pipeline as described by Jace Priester but with a step 6: 6) Support for community add-ons (through python) for preforming basic data/point processing/rendering operations. ParaView provides a large selection of operators, some of which would be trivial to implement in blender if a pipeline was set-up. This idea might also be of interest to Shabbir Marzban, whose recent GSoC email to the mailing list included a proposal to integrate Bundler: 1) Automated rigid structure reconstruction from set of un-ordered images: A very common structure from motion (SFM) tool, Bundler[1] can be used to recover 3D point cloud from such images. Thanks, Isaac Lenton PS: The mailing list seems to be blocking some of my emails, not sure if this will get through. On Mon, 2013-04-29 at 15:02 -0700, Jace Priester wrote: I wrote in point cloud support for Blender over a year ago, but I have been told repeatedly by the developers that there is no interest in the community and therefore they aren't interested in implementing it. Hopefully another voice will help with that. The code isn't extensive as is. It requires a custom data layer to support true per-vertex color (as Blender currently implements vertex color as part of face data - not vertex data). There's a bit of extra code in the mesh drawing functions to access the colors in the custom data layer, and then it draws points to the screen using the set colors. As a bonus, it also supports per-vertex sizing which allows for point distance attenuation (which is necessary for a good appearance). My current code does *not* integrate with any rendering engine - it only works in the editor window - and I gave up trying to get help taking it any further, as I don't use it that often anyway. If a developer knowledgeable enough about BI/cycles wants to hop in and help with finishing this, I'd be happy to collaborate. I see no need for blender to be a point cloud editing program. Meshlab and others are better suited to that, as that is specifically what they are built for. But the ability to import and display point cloud data, in my opinion, is vital to Blender being usable in more fields than it is now. So that said, I think short term and easily attainable goals should be to 1) import point cloud data in a couple common formats (I have done this already) 2) display points with colors and distance attentuation (also done) 3) export point cloud data (also done) 4) some user interface controls for turning colors on/off, adjusting displayed vertex size and attenuation (have not done any of this) 5) support for render at least in BI On Mon, Apr 29, 2013 at 2:19 PM, Antoine Cottin acot...@ultiscan.comwrote: Hi y'all, Sorry for the repost this might be obvious or not, but after re-reading my previous post, I feel that need to mention that our goal is not to produce a tool dedicate to our sole use. Our goal is to develop a tool for us and the Blender community that will be release under GPL. Cheers. Antoine ### My name is Antoine Cottin, I'm a research scientist specialized in active remote sensing applied to earth sciences. For the last 8 years I've been working for/with a laser scanner manufacturer developing processing tools chain for laser scanner data processing. I have an extended knowledge in the various laser scanner systems (hardware and software) and their associated data format. I'm the founder of Ultiscan, a French base company, specialized in laser scanning services. Our main objective/activity at Ultiscan beside laser scanning assets is data fusion between laser scanner data (points cloud) and 3D assets for 3D rendering and animation. Here is an example of one of our last project. This purpose of this project was to merge a 3D asset (the front gate of the heritage building) with a point cloud of the whole building to show to the client how the restoration of gate of the building would look like (some work on texturing is still required): https://www.dropbox.com/s/u6bj80fi8ajqq8r/00_Gruss_Fusion_100p_Black.mp4 The gate have been mesh, from the laser scanner data, and then retopologize (from ~2500k faces to 600 faces), texture and render with Blender. The points cloud animation was done with a commercial software. Then the two pieces was merged together. To do that, we used Blender for the camera tracking, the render and the compositing of the final
