I have seen that researchers are adding some functionalities to the original version of Caffe and use those layers and functionalities according to what they need and then these versions are shared through Github. If I am not mistaken, there are two ways: 1) by recompiling Caffe after adding c++ and Cuda versions of layers. 2) writing a python code for the functionality and call it as python layer in Caffe.
I want to add a new layer to Caffe based on my research problem. I really do not from which point should I start writing the new layer and which steps I should consider.
My questions are:
1) Is there any documentation or any learning materials that I can use it for writing the layer?
2) Which way of above-mentioned methods of adding a new layer is preferred?
I really appreciate any help and guidance
Thanks a lot
For research purposes, for "playing around", it is usually more convenient to write a python layer: saves you the hustle of compiling etc.
You can find a short tutorial on "Python" layer here.
On the other hand, if you want better performance you should write a native c++ code for your layer.
You can find a short explanation about it here.
Related
I’m new to deep learning and I was reading some state of art papers and I found that mask r-cnn is utterly used in segmentation and classification of images. I would like to apply it to my MSc project but I got some questions that you may be able to answer. I apologize if this isn’t the right place to do it.
First, I would like to know what are the best strategy to get the annotations. It seems kind of labor intensive and I’m not understanding if there is any easy way. Following that, I want to know if you know any annotation tool for mask r-cnn that generates the binary masks that are manually done by the user.
I hope this can turn into a productive and informative thread so any suggestion, experience would be highly appreciated.
Regards
You can use MASK-RCNN, I recommend it, is a two-stage framework, first you can scan the image and generate areas likely contain an object. And the second stage classifies the proposal drawing bounding boxes.
But the two-big question
how to train a model from scratch? And What happens when we want to
train our own dataset?
You can use annotations downloaded from the internet, or you can start creating your own annotations, this takes a lot of time!
You have tools like:
VIA GGC image annotator
http://www.robots.ox.ac.uk/~vgg/software/via/via_demo.html
it's online and you don't have to download any program. It is the one that I recommend you, save the images in a .json file, and so you can use the class of ballons that comes by default in SAMPLES in the framework MASK R-CNN, you would only have to put your json file and your images and to train your dataset.
But there are always more options, you have labellimg which is also used for annotation and is very well known but save the files in xml, you will have to make a few changes to your Class in python. You also have labelme, labelbox, etc.
I was consulting several references to discover how I may output trained Weka models into Java source code so that I may use the classifiers I am training in actual code for research applications I have been developing.
As I was playing with Weka 3.7, I noticed that while it does output Java code to its main text buffer when use simpler classification (supervised in my case this time) methods such as J48 decision tree, it removes the option (rather, it voids it by removing the ability to checkmark it and fades the text) to output Java code for RandomTree and RandomForest (which are the ones that give me the best performance in my situation).
Note: I am clicking on the "More Options" button and checking "Output source code:".
Does Weka not allow you to output RandomTree or RandomForest as Java code? If so, why? Or if it does and just doesn't put it in the output buffer (since RF is multiple decision trees which I imagine it doesn't want to waste buffer space), how does one go digging up where in the file system Weka outputs java code by default?
Are there any tricks to get Weka to give me my trained RandomForest as Java code? Or is Serialization of the output *.model files my only hope when it comes to RF and RandomTree?
Thanks in advance to those who provide help.
NOTE: (As an addendum to the answer provided below) If you run across a similar situation (requiring you to use your trained classifier/ML model in your code), I recommend following the links posted in the answer that was provided in response to my question. If you do not specifically need the Java code for the RandomForest, as an example, de-serializing the model works quite nicely and fits into Java application code, fulfilling its task as a trained model/hardened algorithm meant to predict future unlabelled instances.
RandomTree and RandomForest can't be output as Java code. I'm not sure for the reasoning why, but they don't implement the "Sourceable" interface.
This explains a little about outputting a classifier as Java code: Link 1
This shows which classifiers can be output as Java code: Link 2
Unfortunately I think the easiest route will be Serialization, although, you could maybe try implementing "Sourceable" for other classifiers on your own.
Another, but perhaps inconvenient solution, would be to use Weka to build the classifier every time you use it. You wouldn't need to load the ".model" file, but you would need to load your training data and relearn the model. Here is a starters guide to building classifiers in your own java code http://weka.wikispaces.com/Use+WEKA+in+your+Java+code.
Solved the problem for myself by turning the output of WEKA's -printTrees option of the RandomForest classifier into Java source code.
http://pielot.org/2015/06/exporting-randomforest-models-to-java-source-code/
Since I am using classifiers with Android, all of the existing options had disadvantages:
shipping Android apps with serialized models didn't reliably work across devices
computing the model on the phone took too much resources
The final code will consist of three classes only: the class with the generated model + two classes to make the classification work.
I'm looking for some simple (free!) entry level software that will allow me to create simple 3d models and export them in a format (JSON?) that can then be read into a webGL programme.
Simple geometry would be a start, then textures would be nice too... I've looked at Blender, and it's just far too advanced for me, and the tutorials I've found are hopeless.
Something simple like sketchup would be good, but afaik you can't export in JSON. I've found some converters that will do .dae to .json, but the ones I've found seem to be for advanced users.
WebGL is new enough that there aren't many packages like this built up around it just yet. That doesn't mean you don't have some options though:
Blender is a good modeler, and if you are willing to put a little bit more time into learning it you can use exporters from Three.js or some others that are around the net. This seems to be the most popular option at the moment.
Unity 3D is more of a scene builder than a modeling app, but it has a lot of ways to get content into it and both J3D and myself have implemented exporters from it.
Maya is a great modeling tool if you have a way to get access to it (it's commercial), and has Inka to get WebGL content out.
If you want to use something like SketchUp, it should be able to export to COLLADA, which can then be imported into Blender/Unity/What have you and exported from there using one of the previous methods.
As far as formats go, there's no real standards just yet. Most of the exporters will spit out JSON, mine uses a mix of JSON and Binary for speed/size, and some will actually give you Javascript code to execute. Which format to use probably depends on what you want to do with it. I encourage you to experiment with several and see what you like and what you don't.
I'm looking to play around with the Rubinius VM to create a langauage, but just reading the documentation, I'm still quite lost on how to get started. Even looking at the projects, I still can't seem to figure out where the parsing and using the vm comes into place. Does anyone have any resources for this?
Hey I'm a contributor to the Fancy language that runs on rubinius. If you're interested in parsing take a look at boot/rbx-compiler there you'll find a Parser (implemented with KPEG) that basically constructs a tree of AST nodes, each of those nodes has a bytecode method that produces the rubinius vm instructions for everything to work. Fancy share a lot of semantics with ruby, so I guess starting with it would be easy if you're already familiar with ruby. You'll just need to checkout the examples/ dir to geet a feeling on the language and then the kpeg parser, ast nodes, loader, as you progress exploring the compiler. These days Fancy is bootstrapped (that means that the compiler has been written in fancy itself - at lib/compiler) but rbx-compiler is the first step in that process.
Hope exploring Fancy's source code can be of help to you.
In case you hadn't seen it, check out Evan's keynote from 2011 LA Ruby Conf. He shows how to build a simple language, which might be helpful.
There are several packages out there that help in automating the task of writing bindings between C\C++ and other languages.
In my case, I'd like to bind Python, some options for such packages are: SWIG, Boost.Python and Robin.
It seems that the straight forward process is to use these packages to create C\C++ linkable libraries (with mostly static functions) and have the higher language be extended using them.
However, my situation is that I already have a developed working system in C++ therefore plan to embed Python into it so that future development will be in Python.
It's not clear to me how, and if at all possible, to use these packages in helping to extend embedded Python in such a way that the Python code would be able to interact with the various Singleton instances already running in the system, and instantiate C++ classes and interact with them.
What I'm looking for is an insight regarding the design best fitted for this situation.
Boost.python lets you do a lot of those things right out of the box, especially if you use smart pointers. You can even inherit from C++ classes in Python, then pass instances of those back to your C++ code and have everything still work. My favorite resource on how to do various stuff is this (especially check out the "How To" section): http://wiki.python.org/moin/boost.python/ .
Boost.python is especially good if you're using smart pointers or intrusive pointers, as those translate transparently into PyObject reference counting. Also, it's very good at making factory functions look like Python constructors, which makes for very clean Python APIs.
If you're not using smart pointers, it's still possible to do all the things you want, but you have to mess with various return and lifetime policies, which can give you a headache.
To make it short: There is the modern alternative pybind11.
Long version: I also had to embed python. The C++ Python interface is small so I decided to use the C Api. That turned out to be a nightmare. Exposing classes lets you write tons of complicated boilerplate code. Boost::Python greatly avoids this by using readable interface definitions. However I found that boost lacks a sophisticated documentation and dor some things you still have to call the Python api. Further their build system seems to give people troubles. I cant tell since i use packages provided by the system. Finally I tried the boost python fork pybind11 and have to say that it is really convenient and fixes some shortcomings of boost like the necessity of the use of the Python Api, ability to use lambdas, the lack of an easy comprehensible documentation and automatic exception translation. Further it is header only and does not pull the huge boost dependency on deployment, so I can definitively recommend it.