MLKit provides good API documentations and guides for Face detection (https://developers.google.com/ml-kit/vision/face-detection). However, I can not find any informations about the algorihms/baseline model or related research papers behind the scene. Can someone provide any suggestions about it's implementation ?
It likely uses blazeface for face detection used on mediapipe. I could not find a direct answer, but when analyzing an apk with mlkit face detection blazeface.tfl can be found on assets folder.
Mediapipe pose detection doc mentions blazepose powers the ML Kit Pose Detection API. So, blazeface is likely to power mlkit on device face detection.
Links to documentation, paper, poster,
https://google.github.io/mediapipe/solutions/face_detection.html
https://arxiv.org/abs/1907.05047
https://docs.google.com/presentation/d/1YCtASfnYyZtH-41QvnW5iZxELFnf0MF-pPWSLGj8yjQ/present?slide=id.g5bc8aeffdd_1_0
https://drive.google.com/file/d/1u6aB6wxDY7X2TmeUUKgFydulNtXkb3pu/view
ML Kit's selection of APIs run on-device or in the cloud. The on-device APIs can work when there's no network connection. The cloud-based APIs is built on top of TensorFlow Lite along with the integrated Neural Network API. So we do not need to integrate AI-based algorithms into mobile apps. I found the explanation on wiki:
Face detection can be regarded as a specific case of object-class detection. Face-detection algorithms focus on the detection of frontal human
faces.
A reliable face-detection approach based on the genetic algorithm and
the eigen-face technique.
There are other face recognition algorithms, like Principal Component Analysis(PCA) and Linear Discriminate Analysis (LDA) algorithms.
If I want to investigate an RL algorithm for robotics, how should I be using Gazebo and OpenAI Gym to test, train and benchmark the algorithm? Should I start with OpenAI Gym and take algorithms with good scores into the Gazebo environment for real-world scenarios?
Factors to consider when picking the framework to work within
How much time will it require to get up to speed with whatever you choose?
Will performance in the environment reliably predict performance on an actual robot?
How much will it cost?
If you want to have your approach eventually work on an actual robot, you should be testing in an environment that closely simulates your target environment and platform. OpenAI gym's first party robot simulation environments use MuJuCo, which is not free. Further, these simulations are more for toy control setups than actual robotics problems. You would be better served writing ROS nodes and simulating your problem in Gazebo. You might also look at something like Erle Robotics' gym-gazebo tool which will let you bridge between gym and ROS.
I have few ideas for really simple games that I want to try and code.
I know C/C++/Java/PHP/AS3/Python and few more (and learning a new one isn't really a problem).
I'm looking for a game engine that can give me these features:
cross platform (possibly write once run anywhere solution) simple development env that runs on win 7 (or linux, but I prefer win 7)
I need access to in-app purchases/ads/splash screen and any other way to monetize
facebook/whatsapp integration (simple share options, or invite friends in fb)
ability to connect to database (mySql)
ability to connect to some sort of server (TCP/IP, or just fetch any data from certain URL)
2d graphics, simple lines and shapes drawing, text and numbers writing
most games have timers, so I need good response time when clicking a button or any other place on the screen (i guess that rules out Phonegap)
try to think I want to develop a simple card memory game with 5 seconds timer for each time you try to get a pair.
any ideas?
thanks!
try unreal engine 4 :)
it's editor is amazing
one example project is actually that memory card game :)
you can code in c++ and/or you can use blueprints, which is a visual interface to define your game by connecting boxes in an UML like fashion. Its quite easy even for people with no coding background thus collaborating with non-coders can get quite productive.
looks, usability, tons of features, cross-platform.
its not free, but you can get away with a single month subscription fee, which is 20$ or 25$...
and if you sell your game they get a 5% cut. which is ok I guess.
AS3 is easy to learn and its community is still alive :)
To me it is currently the most exciting cross platform technology out there at that time...
I would also recommend using the GPU accelerated 2D Framework Starling http://gamua.com/starling/
Then use ANE (Adobe Native Extensions) for all platform specific transactions such as Facebook integration or in app purchases.
Check out a simple 2 day project available on web/ios/android with one single code base: http://flappy-turd.com
Good Luck!
I'm trying to build a car which can drive itself using video frames from an Android camera on-board using Neural networks. I haven't bought the hardware (car + android).
Meanwhile, I want to simulate the driver application (on my laptop) by streaming video frames from a 3D simulator and see if works properly.
What's an easy-to-learn and mostly pre-configured engine for this?
I think there are no easy-way to do that. In our robotic laboratory we use ROS to do this type of simulations.
It includes a 3D simulator (Gazebo), camera and sensor simulation and a lot of other stuff. You can see an example of 3D robot/camera simulation in this video.
Unfortunately ROS is not easy-to-learn and there is no complete documentation.
Another possibility is to use Blender (the open source, cross platform suite of tools for 3D). There are some robotic project that use Blender (you can see this page for more info) but I have no personal experience with it.
You can try out V-Rep, even though its a commercial product its much superior in performance. It has built in models for autonomous cars. If you are from India, contact Asimov Robotics, they are the resellers
To update this answer for 2017, Gazebo appears to be one of the leading 3D Simulation engines available today. Definitely worth looking into for this kind of application.
An excerpt from their home page:
Robot simulation is an essential tool in every roboticist's toolbox. A
well-designed simulator makes it possible to rapidly test algorithms,
design robots, perform regression testing, and train AI system using
realistic scenarios. Gazebo offers the ability to accurately and
efficiently simulate populations of robots in complex indoor and
outdoor environments. At your fingertips is a robust physics engine,
high-quality graphics, and convenient programmatic and graphical
interfaces. Best of all, Gazebo is free with a vibrant community.
http://gazebosim.org/
Also, the game Grand Theft Auto has been used to test self driving car algorithms. There are APIs to do everything from change the pedestrian traffic, to lights, to weather and more.
I have to prepare a comparison between the following technologies to present it to my Project Manager, but I fell that I'm lost, so if any one can help I will be thankful
I want to compare between them in the following areas:
the support of online video streaming
the budget of using each one
Learning Time will be needed to learn the technology
Which one is the standard and will target a lot of users
The support if I found any problem
Bugs and security issues
connection to DB, SOA and web services
supporting of multi player
The support of online video streaming
Some of the X3D viewers support video streaming (and some even 3D streaming, for things such as augmented reality).
Which one is the standard and will target a lot of users
X3D is a standardized format, such as JPEG with multiple companies being able to manipulate such data and is even officially recommended by HTML5 specs whereas Unity ties you to a single company. Even if most X3D viewers are plugin-based like Flash, there exists also native implementations such as X3DOM to display/interact with X3D files for any browser that supports WebGL.
Connection to DB, SOA and web services
I would usually recommend using a webservice for interfacing with a DB, and yes, X3D can interact with webservices (XML, JSON). There is even a standard binary format that is fast to transfer and parse large contents faster.
Supporting of multi player
Some X3D-supporting providers offer a multiusers service, such as Bitmanagement's BS Collaborate server, but I've seen people using Darkstar/RedDwarf to make multiusers 3D environments as well.
the support of online video streaming
Unity 3D does not support video streaming, unless done through textures, which will give you a really slow frame rate.
I don't know for sure about X3D, but I would doubt it was really made for such tasks.
Silver light has good video support, it should be easy to stream with.
HTML only supports streaming video if using HTML 5, for which it gives the best user experience when user's browser supports it.
Flash is the de-facto for video streaming. It is extensively widespread. They use it for YouTube for example.
the budget of using each one
The cheapest of them all is HTML, it is free. Then you can theoretically set up something for free in flash using Flex SDK and server streaming technology such as Red5 (both open source and free). After that, I believe that all others would probably be on par cost wise, Unity3D coming in as the cheapest of the paid alternatives.
Learning Time will be needed to learn the technology
Listed in order of fastest one to learn to slowest (assuming no prior experience in any):
HTML
Flash/Silverlight
Unity3D
X3D
Which one is the standard and will target a lot of users
Flash is the most widespread. Its only competitor would be HTML 5, as new browsers tend to support it and its the only possible option on iOS. On the other hand, if 3D is what you want, then Unity3D is the standard for now, might be followed by HTML 5 in the future.
The support if I found any problem
Well, Unity3D would offer you good paid support, flash and silver light also (but only when you pay for streaming server licenses). HTML, X3D will not give you any support, but you can find a lot of information on the internet. There is also extensive information about Flash and Silverlight on the internet, but mostly Flash.
Bugs and security issues
All are pretty secure, I'm just not sure about X3D, but all others are comparable in term of security or bug issues.
connection to DB, SOA and web services
Easy to do with HTML, Flash and Silverlight. Harder with Unity3D, and hardest with X3D.
supporting of multi player
Multi-player what? If you are making a game, then clearly I would say your real options are Unity3D if the game is to be in 3D, Flash if it is to be done in 2D. Check out SmartFoxServer for easy multiplayer server.
My 2 cents:
the support of online video streaming:
Some X3D players do support it. Unity does in some ways : http://unity3d.com/unity/features/audio-and-video
the budget of using each one:
X3D and Unity3d are free. You can pay for Unity licenses for extra features and platforms like iOS and Android. If you need to write plugins for Unity, you'll need the $1500 license. There are no costs for distribution of Unity products.
Learning Time will be needed to learn the technology:
Both X3D and Unity3d have active communities and many online resources and offline books. Unfortunately for X3D, the best content creation tool (Vivaty Studio) is no longer supported officially, but X3D is supported in Maya, Max, Blender, and many other 3D programs. Unity's online docs are excellent and the answers.unity3d.com forum (and other forums) are free and fast.
Which one is the standard and will target a lot of users:
'Standard' Well, HTML is the broadest standard. X3D (if including VRML) is the oldest most widely used 3D standard. HTML you have. HTML5 is coming, 'real soon now' (I'm already turning blue). If you mean 'most readily available' the HTML is #1, Flash is #2 (as everyone has a browser, and most computers come with Flash installed already). Flash needs to be installed. Unity needs to be installed too, but it's at least as fast and easy to install as Flash, and it's gotten millions of downloads, so it's getting pretty pervasive. X3D requires a plugin (this should change sometime 'real soon now' with x3dom on HTML5), but the many X3D players are all a little different from each other.
The support if I found any problem:
All have much online community support. X3D has a spec committee but that's not really support per se, you'd have to contact the X3D plugin provider (Bitmanagement, Cortona, Octaga, Exit Reality, Fraunhoffer, etc.) Unity has great online community forums, you can pay for premium support, but I'd only do that if I needed a serious bug or feature that has no work-around.
Bugs and security issues:
X3D's bugs depend on which player you use. Unity has bugs, but the product is pretty solid (I've only crashed it once, and I use is all day, every day, for over a year). Both have a mind toward security, but neither of these are totally secure, especially since you can write scripts that are inherently not secure. So you have a hand in how secure YOUR content will be. Some X3D players support encryption. Unity products are compiled.
connection to DB, SOA and web services:
You can use something like AJAX or JSON or whatever in all these platforms, no? So if it's by web service, sure. If by direct local access, I know Unity can do that. Both Unity and Flash require cross-server xml files on the server to allow access cross-domain (in the web player for Unity anyway).
supporting of multi player:
Unity has excellent multi player networking components. X3D (spec) supports it too, but it really depends on which X3D player you go with as to how well it actually works. Worst case, you can use AJAX or JSON or whatever to roll your own.
Which you choose depends mostly on what you want to do with it. Flash is generally the best route right now, unless it's all about 3D, then I'd try Unity. But a year from now, HTML5 alternatives will begin to take over. Flash DOES support 3D, there are different ways it can be done. Vivaty had a full-featured X3D player written in Flash, so it can be done. There are several good 3rd party 3d plugins for Flash.
I totally agree with wildpeaks : )
Connection to DB, SOA and web services: easy to do with HTML, Flash and Silverlight. Harder with Unity3D, and hardest with X3D.
Reply: I think X3D is not hardest.
X3D(X3DOM) can interact with webservices (XML) as very easy in this example/tutorial
Flash supports hardware accelerated 3d, and comes out of the box with 3d support. In addition, there is the papervision library for more advanced 3d. Unity3d is also supported
as a flash library.
I would consider Flex as a real alternative to Flash. It has the same actionscript language, but uses a tag based syntax called MXML, similar to silverlight. Database remoting is extremely simple. You can access your .Net/Java/Php objects directly on the front end without having to deal with serialisation issues. All of the Flash libraries are accessible.
There is also the X3D player from instantreality.org, supporting video streaming & decoding, XMLHttp request via scripting and its free for non commercial usage.
Flash 3D isn't good 3D for any application of real-time 3D. It is 2.5D with some tricks.
X3D is easy to learn for simple things and harder as complexity goes up. It does have the advantage of being VRML with pointy brackets so the free content, examples and toolkits are easily found. I did comparison tests of the various players. BS Contact is the best for the ability to handle the most complex content with the fastest frame rate and rich color palette. Network support is still non-standard although XMLHTTP and database connections are easy to bolt on. As others have said, Instant Reality is coming on fast and supported by people with a deep understanding of the past implementations and future requirements.
The decision comes down to the project type. A simple comparison rating such as you are is misleading at best but thanks for giving it a shot. I've used VRML through all of its incarnations and now X3D for world building and now as a source for 3D models in video work in combination with Sony Vegas. For cost-benefit without the need to use very expensive modeling toolkits, it is the best of all the choices.