Trying out fintelia terra for scalable terrain rendering Lucas Walter 2022-09-21 | Looking for open source scalable terrain rendering and found this one written in rust, it streams data from the internet depending on where the viewer is and how close to the ground it is, seems to work fine, with a faster internet connection it should be even better:github.com/fintelia/terraThere must be configuration options to set the date and time, but didn't see them. Ultimately I'd like to use it as a library, overlay geotiffs to get images of a certain resolution and camera intrinsics and position so will probably go back to C++ and osgearth (github.com/gwaldron/osgearth).I commented out the line `.with_fullscreen(Some(winit::window::Fullscreen::Borderless(monitor)))`to get it to be in windowed mode.
MIT Vista simulator ROS node Lucas Walter 2022-06-30 | Interactively drive within stored video/gps/lidar data using ROS. It can't go too far off the original course of the real vehicle, which is understandable.It seems like it ought to be able to go into reverse but doesn't know how to bring in the earlier frames. Also it would be nice if the video could go in a loop and it was smart enough to blend the end and start together.Even nicer but probably much more work if multiple datasets where the vehicle turned down different streets could be commingled, then during playback either route could be steered into.Can it do depth images? Maybe those could be converted into what looks like lidar and it would work with that. If no depth data at all does it still work? Pure odometry but no gps?github.com/lucasw/vista/tree/ros_integration forked from github.com/vista-simulator/vistayoutube.com/watch?v=z0UHdAI3sXIhttps://vista.csail.mit.edu/https://news.mit.edu/2022/researchers-release-open-source-photorealistic-simulator-autonomous-driving-0621
rostopic info/rosnode info text user interface using asciimatics Lucas Walter 2022-01-24 | Just execute this with a running ros system, press 'r' to refresh the topic and node lists github.com/lucasw/topic_state/blob/master/scripts/ros_tui.pygithub.com/peterbrittain/asciimatics
Glowing Grid 360 Lucas Walter 2021-07-05 | Move endlessly over an infinite glowing grid, control the view direction with phone tilt sensors or the mouse.This works fine on my phone but glitches in chrome on the desktop- frame jumps forward and back- maybe it needs more processing time.Panoramic equirectangular version youtu.be/8ADfLafygCk
Glowing Grid Panoramic Lucas Walter 2021-07-05 | Move endlessly over an infinite glowing grid.360 viewer version created with spatialmedia: youtu.be/ZHBaS0F0pIsMade in Blender 2.93 with cycles equirectangular camera and glare composition effect. Tried out 8K frames but ffmpeg locks up with those, so went down to 4K (but youtube only is allowing playback at 2K at least for now?).The render is only 30 frames but is looped 400 times with ffmpeg:ffmpeg -threads 14 -stream_loop 400 -framerate 30 -pattern_type glob -i '*.jpg' grid_pano_long.mp4Try rendering with motion blur, that may reduce aliasing artifacts.Future Eevee versions are supposed to support panoramic rendering, that would have speeded this up a lot, but I like the cycles glare composition effect more than the built-in eevee equivalent.
Starfield 360 Lucas Walter 2020-03-12 | Moving through a panoramic starfield, it can be set to loop and is mostly seamless though the youtube player sometimes stutters.Equirectangular images software rendered with C++ and OpenCV - the stars are cv::ellipses that get wide near the poles at the top and bottom of the frame github.com/lucasw/stars_panoRendered at 4096 x 2048, I would have gone to the youtube maximum of 8192 x 4096 but ffmpeg couldn't handle that - maybe importing as an image sequence into blender would work?3D sequel to the 2D youtube.com/watch?v=oE2ZVhHwxAUgithub.com/lazarus-pkgs/panini was useful for viewing individual images.
time delay map frei0r in gephex Lucas Walter 2019-12-04 | Use a frei0r mixer2 where the first image input is stored in a buffer, and the second is an image where the intensity of each pixel determines how far back in the buffer to get the output pixel. For a vertical gradient this looks just like 'slit-scan'. It takes a lot of moving around of memory for lookups into an array of size width * height * buffer length pixels - maybe storing the pixels with the buffer length as the major axis would be better for cache? Copying the input images into the buffer would then be slowed down...The animated delay map can be seen at 0:33github.com/lucasw/frei0r/blob/delay_map/src/mixer2/delay_map/delay_map.cppgithub.com/lucasw/gephex
galaxy Lucas Walter 2019-11-23 | Using gephex github.com/lucasw/gephexgithub.com/lucasw/binarymillenium/blob/master/gephex_graphs/galaxynew3More graphs github.com/lucasw/binarymillenium/tree/master/gephex_graphsTODO get them running again and re-upload.
frei0r particle edge glow Lucas Walter 2019-11-17 | Using frei0rs with ROS github.com/lucasw/frei0r_image and frei0rs from github.com/dyne/frei0r, also using github.com/lucasw/image_manip and github.com/lucasw/signal_generate
ros drag-and-drop image publisher with imgui Lucas Walter 2019-04-19 | github.com/lucasw/imgui_rosDrag and drop images onto an imgui window, it then is published as a ros Image to rqt_image_view.Using SDL2 which gets the drop events in the form of file names, which are then loaded from disk with opencv. Would like to be able to received copied images from Chrome, but those are different (the entire image is on the clipboard?).Went back to ros1 with imgui_ros because of ros2 uses too much cpu, haven't ported all the 3D visualization parts yet.github.com/lucasw/imgui_ros/issues/81
Starfield parallax side scrolling Lucas Walter 2019-02-12 | github.com/lucasw/love2d_test'endless' stars moving to the left- there ought to be something better (and longer than a few seconds) already on youtube but it seemed faster to make one from scratch than dig deep into youtube search results. (There are a lot of moving forward through stars, but not the side scrolling)Stars are just circles, should try making a few be made out of circles with an alpha gradientAlso could have bigger stars with lens flares, also via a transparent image rather than a drawn circle.The drawn circles probably take more cpu than a scaled down image.Recorded with simplescreenrecorder in Ubuntu.Panoramic sequel youtube.com/watch?v=qrQOVp3ivhYUsed in youtube.com/watch?v=8Q_PtjvVLt4
polar projection omnidirectional - opengl + imgui Lucas Walter 2019-01-08 | Same as youtube.com/watch?v=Ux1Tmk33l8o but with a polar projection- look straight a flat plane at it appears as a circle. The projection loops over itself near the edges.
cube map omni-directional camera- OpenGL, imgui, ROS2 Lucas Walter 2019-01-05 | Use render to cube map then a lens geometry is used which has normals that are used to lookup into the cube map in the shader. More bizarre and complicate lenses are possible.ROS tf is used for the positioning of geometry - if not for the imgui widget the position could come from another tf publisher, and the cube image is getting published out as a ros sensor_msgs Image.github.com/lucasw/imgui_ros/issues/52See also youtube.com/watch?v=t3fQJTo7QWc
ros2 + imgui + usb_cam - stop motion animation Lucas Walter 2018-11-15 | Stop motion image capture pipeline and ui with ros2 and imgui and usb_cam.Not sure what k2so is doing- dancing?Also though it would be easier to concatenate an mp4 from kazam with an image sequence of a different resolution in shotcut, but I think shotcut is confused about framerate and resolution.Sequel to the ros1 version: youtube.com/watch?v=mc1xC_bGN5Egithub.com/lucasw/image_manip/blob/bouncy/image_manip/launch/stop_motion_launch.pyros2 version of usb_cam:github.com/lucasw/usb_cam/tree/bouncy
imgui + ros2 + rotozoom Lucas Walter 2018-11-15 | The imgui interface is itself a ros2 node which is configured via service calls to create Image displays and topic publishers connected to sliders (floats and ints here).github.com/lucasw/imgui_ros/tree/bouncy - currently using the SDL + OpenGL3 backendOld ros1 qt version: youtube.com/watch?v=2vLw7mZvy1Mgithub.com/lucasw/image_manip/tree/bouncygithub.com/ocornut/imgui
Lambda sound wave 2D simulation Lucas Walter 2018-10-04 | I've taken the lambda code and turned it into a library and a ROS node that sets up and executes a simulation.github.com/lucasw/lambda_synthIn the upper left is a positive reflecting wall, on the right a negative reflecting wall, and an absorbing wall on the lower left (and what does a regular wall in real life correspond to?).It doesn't really use ROS at this point but costmaps could be used as walls later on. Being able to move the walls during the simulation is a goal but maybe requires deeper digging into lambda to support.
Open Pseudo Spectral Time Domain sound simulation Lucas Walter 2018-09-30 | Trying out openpstd. This is similar to lambda sim in that there are audio sources (speakers) and sinks (microphones), and sound waves propagate around the room in a very cpu intensive way.The 'pseudo spectral' method is supposed to be faster than the more traditional technique used in lambda, but this non-apples-apples example actually took a much longer time to calculate than a similarly simple room in lambda. Perhaps it is higher fidelity than the lambda equivalent.Are diagonal walls supposed to be possible?It isn't clear there is any way to export to audio or even raw floating point files for either the microphone or the entire scene. png/jpg export didn't work correctly (or my settings were wrong). It wouldn't be hard to hack support for this in, that part of the code looks straight-forward enough.This is ~2.0 built from the master branch (with a few modifications to make it run work with Ubuntu 18.04 - see github.com/lucasw/openPSTD/tree/include_vector and gist.github.com/lucasw/574e325be0c4ccf7112e3d8f68caa954 ).So far I can only run the default room which is in this video- any modifications result in a crash. It isn't clear how to modify the domain (room walls), some are clearly set to reflect and others to absorb the sound waves. I'm not sure what the number labels on the walls mean.There is 3.0 and 3.1 branch but those don't work at all for me, both crash immediately after attempting to simulate and take 100% cpu even when doing nothing.I think I'll go back to lambda, which is more robust and looks easy enough to modify for my purposes. I also have some old electromagnetic wave propagation code lying around that could probably be modified to simulate sound instead.github.com/openPSTD/openPSTDhttp://www.openpstd.org
spectrogram audio paint Lucas Walter 2018-09-10 | There are some really irritating tones in a few sections I'll try to edit out. Or it's all irritating tones and noises, start with a low volume and proceed at your own risk.Slower speed playback avoids some of that.Paint spectrograms in Gimp and exported tifs are turned into looping audio.It takes a second or two from when I export the image (which may be any number of seconds after the image is edited) before the changes take effect and are heard.Kazam set to capture the window only doesn't get the menus when I use them, but most of the effects ought to be self explanatory.Use the area capture next time.Used Gimp 2.9 to get 32-bit floating point images, but probably regular lower precision images work equally well.github.com/lucasw/spectrogram_paint9:00 - sort of neat frequency slide19:30 - some nice pulsing noise25:10 - noise + slide thing
Tao physical model sound synth - 40 strings Lucas Walter 2018-08-23 | This is the example of a simulated 40 string instrument being 'played' in taopm.I've imported the Tao 'A Software Package for Sound Synthesis with Physical Model' (http://taopm.sourceforge.net) into github github.com/lucasw/tao_synth, replaced configure with cmake, and eliminated what I don't want to maintain. Some of that may come back as python scripts.The visuals and sound don't match up, the visuals are proceeding at a tiny fraction of the audio rate.add and re-encode audio to silent ogv videoffmpeg -i 40strings.ogv -i 40strings_b.mp3 -vcodec copy 40strings_with_audio.ogv
bullet physics soft body robotic grasping Lucas Walter 2017-10-04 | Rigid body grasping on soft body pressurized sphere. The sphere is regenerated with different pressures and later different surface tesselations.roslaunch bullet_server bullet_server.launch use_sim_clock:=true rigid_only:=falseroslaunch bullet_server grasp.launchrosrun bullet_server icosphere.py __ns:=bullet_server _pressure:=500 _mass:=1.5 _levels:=3It would be good to try a volume sphere as well, or turn it into an ellipsoid or make other shapes.Perhaps the friction could be cranked up so the squishier spheres wouldn't slip out. Maybe finger tips would help too.Picking up a piece of simulated cloth would be interesting, and probably difficult.bullet_server is using about 60% of a cpu with about 3000Hz internal update rate, that could probably be reduced (though soft body penetrations seem to get more common then), and the real time factor increased for faster simulation.I'm using an rviz branch that allows light sources, the bullet_server.launch that launches rviz will likely fail elsewhere (so just start a new one and subscribe a Marker to visualization_marker_array).
grasper - Bullet Physics Library in ROS rviz Lucas Walter 2017-09-28 | Inspired by the 'open sandbox for training robotic hands to grasp things' talk from ROSCon 2017. Next thing to do is make the ball a squishy soft body instead of the current hard 'billiard' ball.github.com/lucasw/simple_sim_rosroslaunch bullet_server bullet_server.launchroslaunch bullet_server grasp.launchrosrun bullet_server grasp_sequence.pyAlso requires pid package.
RotoZoom ROS Nodelet Lucas Walter 2017-07-09 | Rotate and zoom a ros image in a nodelet.github.com/lucasw/image_manipIt would be nice if the rqt dynamic reconfigure could have a mode to send constant updates rather than waiting for the slider to be released.
IIR Video Filtering ROS Nodelet Lucas Walter 2017-07-08 | Apply an infinite impulse response filter to a video stream. Not that interesting until a large a coefficient is applied.The video gets fouled up with large coefficients because the processing is being done with 8-bit color- probably the frames should be filtered in 16-bit signed and then converted back to 8 for published, though with a large performance penalty.github.com/lucasw/image_manipSource video is the cc licensed youtube.com/watch?v=ERtQIyaw5Yg
OpenSimplex noise image in ROS nodelet Lucas Walter 2017-06-26 | Generate a noise image in a ros nodelet, dynamic reconfigure to change the noise scale factor for different 'octaves'.github.com/lucasw/open_simplex_noise_rosSort of sequel to old vimjay version youtube.com/watch?v=JuRLf9DHGrw
Atari 2600 example with ca65 assembler Lucas Walter 2017-05-03 | github.com/lucasw/atari2600_ca65_exampleNot really pushing the (emulated) hardware much, just learning the basics of the VCS and 6503 assembly.It would be nice to have more levels, use additional playfields, and maybe multiple gaps in the same wall occasionally, and scrolling stars in the background using the missile sprites.The stella emulator looks to be oriented around DASM rather than ca65, it doesn't load the label and map outputs generated by it, maybe that could be added easily.
ROS bullet sim tracked vehicle Lucas Walter 2016-12-12 | This is a full physical simulation of a tracked vehicle.Each wheel has gear teeth which intermesh with the gaps between the track plates.There isn't any suspension, no tensioning between the drive wheels, and the friction coefficients might be too low, which results in a poor ability to climb up obstacles.There is also a bug in bullet_server where fixed constraints aren't being handled properly, so I've used hinges instead with small angular limits.github.com/lucasw/simple_sim_rosalso clone github.com/mogumbo/rviz_lightingsudo apt install ros-melodic-pid then build and run:roslaunch bullet_server bullet_server.launchrosrun bullet_server ground_plane.py _ns:=bullet_serverroslaunch bullet_server tracked_vehicle.launch
Spawn sim rigid bodies with rviz interactive markers Lucas Walter 2016-12-06 | The multiple overlapping markers make for a clunky interface, but it is still nice to be able to do this.github.com/lucasw/simple_sim_ros/tree/softbodyroslaunch bullet_server bullet_server.launchroslaunch bullet_server spawn.launch
Soft body simulation in bullet physics + ROS with rviz Lucas Walter 2016-11-20 | A python script generate a rigid body chassis anchored to cube soft body wheels, and then motors on hinge joints propels the vehicle forward.Spoiler alert: it turns out square squishy wheels don't work very well.github.com/lucasw/simple_sim_ros/tree/softbody
Rviz Interactive Marker control of tf frames Lucas Walter 2016-11-04 | github.com/lucasw/rviz_interactive_tfAlso youtube.com/watch?v=C9BbFv-C9Zo has something similar implemented.
Rviz Textured Quad Lucas Walter 2016-10-27 | Use an image topic as a textured quad in rviz.github.com/lucasw/rviz_textured_quads/tree/image_topic
Stewart Platform Simulation in ROS Lucas Walter 2016-10-17 | Stewart Platform simulation in ROS using a light wrapper around bullet physics (github.com/lucasw/simple_sim_ros). Currently the actuators only have velocity control, which I'm setting to positive or negative values in rqt message publisher.The system is buggy and locks up around 1:00, unfortunately.
Pop can challenge simulation retrieval fail Lucas Walter 2016-09-06 | github.com/lucasw/popcanbothttp://www.robothon.org/robothon/popcan.php
ROS bullet server Lucas Walter 2016-05-05 | A thin ROS wrapper for bullet physics (http://bulletphysics.org), uses rviz for visualization. Bullet already is used within Gazebo, but many advanced features of bullet are not- soft bodies especially (though I'm not supporting soft bodies yet either).github.com/lucasw/simple_sim_rosBodies and constraints (joints) can be created live, which is not always a great idea for stability but can be entertaining- when a constraint is created between two distant objects they sometimes spring high into the air.Bodies and Constraints and Impulses are created via topics currently, though at least for the first two I should switch to services.
ROS + OpenCV image distortion and rectification Lucas Walter 2016-03-07 | This shows off code to distort an entire image according to OpenCV fx,fy,cx,cy and distortion coefficients, and then undistort it using the standard rectification function. The source image is on the left, the distorted image next, then the rectified image, then the difference of the input image and the rectified image on the right.The difference isn't always zero because information is lost in the distortion- parts of the image go out of the boundaries, or are so distorted to a subpixel level they can't be brought back.I've been meaning to git forward image distortion added to OpenCV for several years, but have neglected to push for it. The source code is currently here github.com/lucasw/vimjay/blob/master/src/standalone/distort_image.cppUsing ros:roslaunch vimjay distort.launch
ROS Stop Motion Animation Tools Lucas Walter 2016-01-16 | Lego stop motion the Robot Operating System and my nodes from github.com/lucasw/vimjay and github.com/lucasw/v4l2ucpMore details here:http://lucasw.github.io/ros-stop-motion-animation
nngen Lucas Walter 2015-03-10 | Experimenting with graph based visual generation with graphs that are built automatically rather than under user control. This isn't that interesting past about 30 seconds, but I left it rendering and ended up with 10 minutes worth. The next iteration needs to generate more variety, this is just random linear combinations of basis images, and also do more interesting things on the time axis than move things up and down or left and right.github.com/lucasw/vimjay/blob/master/src/nngen/nngen.cpp
Lane Finding with OpenCV Lucas Walter 2015-01-27 | This isn't very robust yet, other lighting conditions introduce a lot of noise that mess it up (though shouldn't be too hard to improve), and lane changing and other cars aren't handled at all.The source code is available under the GPLv3 license github.com/lucasw/road_vision
OpenCL live coding Lucas Walter 2014-07-26 | Live coding with OpenCLModifying a simple filter on webcam input. I make syntax mistake at one point, the error is shown in the lower left terminal.The GPLv3 source code is here:github.com/lucasw/cl_misc
Pixbot - javascript pixelated turtle graphics Lucas Walter 2014-02-11 | Queue up commands to a turtle cursor that draws on the screen. Select commands from the left-hand panel, and they appear along the command queue on the bottom. Press go to execute them.You can try it out here: http://lucasw.github.io/pixbotOr download the GPL 3.0 project here github.com/lucasw/pixbotInspired by http://en.wikipedia.org/wiki/Turtle_graphics and http://light-bot.com/hoc.html from the Hour of Code.
Down and Out in the Magic Kingdom - text to speech audio Lucas Walter 2014-02-06 | Cory Doctorow's Down and Out in the Magic Kingdom converted to an audio book plus images of the words using pico2wave on linux, imagemagick convert for images, and avconv/ffmpeg to generate the videos and concatenate them all togethere. The details are here http://binarymillenium.com/2014/02/text-to-speech-audio-books-with-text.htmlThe generous Creative Commons license that allows derivatives makes this possible/legal, the book is available here:http://craphound.com/down/?page_id=1625There is already a much more understandable human generated audio book version of this elsewhere on youtube, but I thought it would be fun to try this process out.
screengrab ROS node Lucas Walter 2013-11-30 | ROS node for capturing from the screen, with ros parameters for setting the x,y,width, & height of the region of interest on the screen.github.com/lucasw/screengrab_rosSome other discussion is here: plus.google.com/+LucasWalter/posts/f7fLhyWL3A4
GPS tagged video from 2004 Lucas Walter 2013-10-24 | OpenGL visualization of a laptop recorded serial-over-usb data from a TripNav GPS magnetically attached to car roof + Canon DV camera pointed out the passenger side window. This is driving from Ballard to SoDo.
Google Ceres solver for bezier curve around obstacles Lucas Walter 2013-09-10 | Learning to use google ceres solver (code.google.com/p/ceres-solver/). An initial bezier curve may pass through some randomly placed obstacles and then the solver tries to both minimize the error resulting from intersecting an obstacle and minimize the overall path length by moving the bezier control points around.The white line is the initial curve, the green line is the solution Ceres found, and the grey squares are the obstacles. The control points also have lines drawn between them.It doesn't always work that well, and some refinements could probably improve it, and also using bezier control points probably isn't as good as just moving the path itself around directly (though the curve smoother).The source code to this is GPL and located here: github.com/lucasw/bezier_solveThe music is Kevin MacLeod 'Blipotron' and is Royalty-free/Creative Commons Attribution. See http://incompetech.com
A-Star Search Lucas Walter 2013-08-21 | A* or A-star search implemented in Processing/Java. The 'maze' is randomly generated as are the start and goal points.Most of the work was from 2008 but I hadn't until now made any videos of it, the source code is available on github.com/lucasw/binarymillenium/blob/master/processing/astar/astar.pde (There used to be an embedded applet linked from code.google.com/p/binarymillenium/wiki/AStarSearch before google archived google code)I think there are cases where it won't choose a slightly shorter path but I may have gotten most of those.Music is the creative commons licensed free for commercial use Kevin Macleod - Cut and Run http://incompetech.com/
Kevin MacLeod Compilation/Sampler Lucas Walter 2013-07-25 | The music and titles get unsynchronized especially near the end, I'll try to see if I can fix that.I've been seeing Kevin MacLeod credited all over the place in the last year, in indie games and random youtube videos (especially monetized ones) and elsewhere, but only recently looked him up. The next step seemed to be to start using his music in my own stuff, and first I would have to listen to a lot of it and get familiar with it. I then downloaded a bunch of mp3 and the set about trying to turn a folder of mp3s into videos with static title images, and then joining those together into one long video with a script.The latter step was a struggle until I also came across a partial explanation of the the whole acrimonious ffmpeg/avconv split, and then promptly went back to using a proper version of ffmpeg instead of the Ubuntu supplied avconv.Kevin MacLeod royalty free Creative Commons Attribution music from http://incompetech.com/:8bit Dungeon Boss8bit Dungeon LevelAmblerBabylonBasement FloorBlipotronBlue ScorpionCall to AdventureCephalopodChipper DoodleChipper Doodle v2Cipher2Clenched TeethClub DiverCognitive DissonanceConsequenceConstancy Part ThreeConstancy Part TwoCut and RunCut TranceThe scripts for making the static image music videos are here:http://binarymillenium.com/2013/07/turn-set-of-mp3s-into-static-image.html
Vimjay - videos from live set Lucas Walter 2013-07-13 | Visuals created with vimjay: github.com/lucasw/vimjayThe music is the Creative Commons licensed Perpetual Motion by destinazione_altrove http://ccmixter.org/files/destinazione_altrove/42765I need to make it easier to have a variety of stock saved graphs, and be able to quickly load them while modifying their source imagery. Working up a novel graph (or rough replication of an old one from memory) is sort of interesting as performance art but not as suitable for a dance floor.Also having a default bank of source imagery videos to switch among, or cycle through- right now there is one directory of random bitmaps, and some number images- those got repetitive quickly.Rot2D is overused, I need to use it more subtly- very rarely rotate or zoom out beyond certain thresholds that make it obvious that effect is in use.It needs to be easier to create instrument panels for graphs, so that navigating the whole node graph in the standard ui is not necessary while just making numerical modifications.
soundpaint - mouse synth Lucas Walter 2013-05-24 | Use the mouse to draw a sound waveform (in red), then different keys play the sample (repeated back to back about 10 times) at different sample rates. It is also possible to draw on the frequency and phase and update the waveform.github.com/lucasw/soundpainthttp://binarymillenium.com/2013/05/soundpaint.html
Voxel paint - keyboard controlled 3d voxel editor Lucas Walter 2013-04-20 | The mouse is used to rotate the voxel preview, and it can be used to paint also but the keyboard and vi-keys are the primary interface. It doesn't export in any format other than a sequence of 2D images that are slices of the voxel model. (What format does Minecraft use?)I'm using processing with a separate P3D graphics target for the voxel preview, but it seems not to work completely- ortho() didn't work and neither did changing the fov or trying to change lighting.To run this you will want to download Processing (http://processing.org/download) and my code which is here (and free and open source):github.com/lucasw/pixelpaint
