I see that CIDEr is a universal metric in Image Captioning.
I want to use this metric on my project, but I can't find any library. There is few info. CIDEr can only be seen in paper. And I can have searched just BLUE library which is used to evaluate NLP.
Anybody knows how to use this metric?
If there is no way, how can I evaluate my image-captioning model?
p.s. I use OFA model on my project.
https://github.com/tylin/coco-caption
Here is the github link which can be the answer. I can use how to apply CIDEr metric on my project by referring this repository.
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 have spent the days looking for an Opensource JavaDoc editor or a way to edit JavaDocs easily or creating package-info files and come back empty.
I am basically after a way to change the way the packages are listed. The automation framework I work on is undergoing some potential future changes. These changes in a nutshell will organise our test scripts in a way where they are split up into business areas and then into smaller packages of business functions within that area.
We use IntelliJ and the default JavaDoc generator creates the JavaDocs in a way where it makes a list of all the packages, what I want to do is have it create a list of the main packages (business areas) and then the sub packages within that contain our test scripts (business functions).
For example:
BusinessAreaOne contain BusinessFunctionOne, BusinessFunctionTwo and BusinessFunctionThree.
BusinessAreaTwo Contains BusinessFunctionFour and BusinessFunctionFive.
The BusinessFunction packages in turn contain scripts that cover just that business function.
I haven't been able to find an editor that does what I would like and I don't think editing the files manually is practical each time there is an update.
If there is no opensource java editor can do this are there any suggestions of a work around?
As I mentioned I have tried creating package-info files but as far as I can tell you can only add a description to that package. Is there any tutorials out there or does anybody know a way I can get it to display the list of packages within the selected package?
Thanks in advance.
Javadoc doesn't treat package hierarchies as nested hierarchies, it just flattens them all into a list, as you probably noticed, so your business functions will just appear as entries in the list between the business areas.
Looking for a Javadoc editor is not going to help you because the Javadocs you end up with is the output from the Javadoc compilation process, and there's no specific tag you can put in your code that will influence the compilation process in the way you're looking for.
You would have to write something to override Javadoc's implementation of its HTML formatter specifically for the package list using:
Javadoc Doclets
You can customize the content and format of the Javadoc tool's output by using doclets. The Javadoc tool has a default "built-in" doclet, called the standard doclet, that generates HTML-formatted API documentation. You can modify or subclass the standard doclet, or write your own doclet to generate HTML, XML, MIF, RTF or whatever output format you'd like. Information about doclets and their use is at the following locations:
Javadoc Doclets
You would use your own implementation of the Standard Doclet, finding the right class to override - try looking first at the PackageDoc class. I'm going through this learning process myself at the moment, but here's the info on Doclet and there's a link at the bottom of that page to the source code which will illustrate how to do it.
Of course there could be an easier way of doing it and if I discover it I'll come back to update this answer.
We have a group of developers moving from C++ to C# and WinRT. We used D'Oxygen as part of our C++ developer builds, and I'd like to continue to have document generation as part of the developer build in C#/WinRT.
It's easy to turn on XML Doc generation, and I believe that will provide warnings for malformed tags, but without actual HTML output, I think our developers will be missing valuable feedback.
Looks like NDoc is now defunct, and I took a quick look at Sandcastle, but found it rather complex. Ideally, I'm looking for something that doesn't unduly burden developers, or require them to remember extra steps as they edit, build, test, and commit. In other words, the best solution would be something that "just happens", like a post-build step, and doesn't add significantly to each developer's build time.
If anyone has had some experience doing this in C#/WinRT, I'd sure like some advice.
Thanks in advance!
Get Sandcastle Help File Builder.
Create a help project for your library in the Visual Studio solution.
Remove Build check mark from Debug solution configuration to build the documentation project only in Release configurations, since Debug is most often used during development. For release build testing or performance testing you can either create another solution configuration or simply switch the option back and forth.
Build the documentation once
Include the documentation file in the solution so it shows up in the Pending Changes window when the file changes.
Kindly ask your developers to build with the release configuration that updates the documentation before check-in or use any other policy to require updating the documentation.
I don't think it makes sense to build the documentation all the time, but it helps to make it easy to do so that when you actually need an updated version - you can build it really quickly.
You can also make sure to use FXCop or StyleCop (forgot which) and configure it to treat missing XML documentation warnings as errors - at least in release builds. Doing it for debug configurations might slow down development and make changes difficult since developers often want to try things out before committing to a final implementation worth documenting.
EDIT*
Sandcastle provides various output formats as shown in the project properties:
I would like to mention ForgeDoc (of which I'm the developer), it could be what you are looking for. It is designed to be fast and simple, and it generates proper MSDN-like HTML output. It also has a command-line interface so you can just call it from a post-build event command in Visual Studio.
I think you should give it a try, as I would really like to hear about your thoughts.
Having recently read "Growing OO systems guided by tests", I am very impressed with the windowlicker testing utility for java/junit. Basically it wraps the GUI and GUI-interaction with drivers and gestures, so your integration/end-to-end tests can be written neatly like:
//setup
ui.enterUserDetailsFor(newUser)
ui.sendForm()
//assert
ui.showsWelcomeMessage()
All swing-gui-thread synchronization and finding widgets etc is nicely isolated in the framework. Exactly this sort of higher level testing is something I really miss in my current wxPython project. Is there anything at all similar for wxgtk/wxpython?
What I can tell from the windowlicker source it builds on java.awt.robot, if there isn't a windowlicker, is there anything like robot?
I haven't heard of anything specifically for wxPython. You can use the Widget Inspection Tool for some of the stuff you are talking about though. See here:
http://wiki.wxpython.org/Widget%20Inspection%20Tool
There's also the Sikuli project which is GUI-agnostic:
http://groups.csail.mit.edu/uid/sikuli/
Hopefully one of those will help you out.
Inside a Java project I use Google Protocol Buffers (GPB) for serializing my objects. I can use the same .proto files in auxiliary Python code, which is great. Now I'm adding a Flex client to the whole thing and I'd like to use the same .proto files once more.
It seems there's a couple of projects out there which compile .proto files to Actionscript. From a few glances at the projects' homepages, it seems to me that protobuf-actionscript3 is actually the most advanced and most "alive" of these projects.
Has anybody had practical experience with GPB to AS3 compilers and which one(s) can you recommend (or recommend against)?
If you're sure you want to use GPB, then protobuf-actionscript3 is your best option. It builds on the semi-successful protocol-buffers-actionscript project: http://code.google.com/p/protocol-buffers-actionscript/
If you're open to looking at other formats, there's always Adobe's own AMF3. It seems to have a good amount of community support behind it.
The only choice now is https://code.google.com/p/protoc-gen-as3/. All the other Protobuf/AS3 projects are out-of-date, and lack of features.