Nulab is the online project management solution for business &amp; software development. Try our collaboration tools to boost teamwork &amp; achieve product goals.

Collaboration Tools for Team & Project Management | Nulab

Nulab is a team of fun-spirited creators dedicated to creating software the improves team collaboration.

Nulab - Crunchbase Company Profile & Funding

The Nulab team is dedicated to making products for passionate creators &amp; collaborators like us! Learn more about our company culture &amp; multinational community.

Company & Leadership | Nulab

Nulab | 9,179 followers on LinkedIn. You&#x27;ve got the vision, we&#x27;ve got the tools. Online collaboration tools for modern teams. | Nulab is a growing software company that provides online collaboration tools for teams of all sizes. Our products Cacoo, Backlog and Typetalk help 2.5 million global ...

Nulab | LinkedIn

We are the creators of Cacoo, Backlog, and Typetalk—online collaboration tools that make your work life simpler. Our suite of project management and communication software helps teams keep track of work. Collaborate with the whole crew in real-time for faster feedback and quicker turnaround times.

Nulab

Information on stock, financials, earnings, subsidiaries, investors, and executives for Nulab. Use the PitchBook Platform to explore the full profile.

Nulab Company Profile: Stock Performance & Earnings | PitchBook

Customers shall not assign or otherwise transfer their status under these Terms or any rights or obligations under the Terms to a third party without Nulab’s prior written consent. However, if a Customer undergoes a merger, etc. resulting in a comprehensive succession, the surviving company ...

Terms Of Service

Nulab - Cloud-based collaboration tools. Public Company. Raised a total funding of $5M over 1 round from 3 investors.

Nulab - Company Profile - Tracxn

We are a small company with a big heart and all of our products are handcrafted with love. - Nulab Inc

Nulab Inc · GitHub

Nulab Corporate Headquarters, Office Locations and Addresses | Craft.co

Nulab Corporate Headquarters, Office Locations and Addresses | ...

Discover the benefits of working at a multinational company with diverse culture, work-life balance, &amp; hybrid or remote work options. Intrapreneurs welcome.

Job Openings & Career Support at Nulab

We’re sorry but our site doesn’t work properly without JavaScript enabled. Please enable it to continue · Loading

Nulab NYC Office: Careers, Perks + Culture

See the company profile for Nulab Inc. (5033.T) including business summary, industry/sector information, number of employees, business summary, corporate governance, key executives and their compensation.

Nulab Inc. (5033.T) Company Profile & Facts - Yahoo Finance

Dun &amp; Bradstreet helps companies improve their business performance through data and insights delivered through our Data Cloud and Live Business Identity

Nulab, Inc. Company Profile | Clearwater, FL - Dun & Bradstreet

Buy Nulab branded products including t-shirts, tote bags, hoodies, and stickers from our new NuStore! Plus swag from our products.

Grab some swag at Nulab's new NuStore! | Nulab

Cacoo is an online diagramming tool that lets you easily create wireframes, flow charts, UMLs, etc

How to compose independently computed Kotlin nullables, in an easy and clean way In Kotlin we have . null safety Despite nullable , there are many advantages to using it. Your API can express in a clean/compile-time-safe way, and bring back your function to the world. can’t be totally considered as a monad optionality side-effectful pure I find this feature as one of the most exciting and idiomatic of this language. Anyway, there is one drawback. Composability! I will propose a solution to this annoyance. Have you ever tried to compose nullables? Well… I can’t say that composing independently computed values really shines. Let’s figure out a real-world use-case. Crypto exchange Let’s think about a crypto exchange. Suppose that when registering, they ask you for your name, a username, and an email. Once registered you can start crawling the crypto world. You get excited about this world, and you want to start buying crypto. You can’t! Before you have to fill in your phone number, your credit card and complete the . All these are optional info, related to the user. KYC Let’s model it in our business logic: ( username: String, name: PersonalName, email: Email, phoneNumber: PhoneNumber? = , creditCard: CreditCard? = , kycVerification: KycVerificationData? = ) data class CryptoUser val val val val null val null val null The exchange has the procedure that, in order to let you buy crypto, requires those optional data. becomeRich ( username: String, phoneNumber: PhoneNumber, creditCard: CreditCard, kycVerification: KycVerificationData) = TODO( ) data class BuyCryptoInfo val val val val fun becomeRich (crypto: , buyInfo: ) CryptoInfo BuyCryptoInfo "conquer the world" Non-null from nullables Can you figure out a way to go from to ? CryptoUser BuyCryptoInfo You have two options, proposed as the following factory methods: ( username: String, phoneNumber: PhoneNumber, creditCard: CreditCard, kycVerification: KycVerificationData){ { : BuyCryptoInfo? = with(user){ (phoneNumber != && creditCard != && kycVerification != ) 
 BuyCryptoInfo(username, phoneNumber, creditCard, kycVerification) } : BuyCryptoInfo? = with(user){
 phoneNumber?.let {
 creditCard?.let {
 kycVerification?.let { 
 BuyCryptoInfo(username, phoneNumber, creditCard, kycVerification) 
 }}}
 }
 } 
} data class BuyCryptoInfo val val val val companion object fun from1 (user: ) CryptoUser if null null null else null fun from2 (user: ) CryptoUser Personally, I don’t like any of those solutions. But let’s continue. You click on the “ ” button, and suddenly you get… . buy Something mandatory is missing, please check your info With , we are only telling that something is missing but not what. Let’s change the code! nulls sealed {
 data {

 companion object{
 fun from1(user: CryptoUser): Result<BuyCryptoInfo> = (user){ (phoneNumber == ) Result.Error( ) (creditCard == ) Result.Error( ) (kycVerification == ) Result.Error( ) Result.Ok(BuyCryptoInfo(username, phoneNumber, creditCard, kycVerification))
 }
 } 
} < > class Result out T < >( : ): < >() ( : ): < >()
} ( : , : , : , : ) class Ok out T val value T Result T data class Error val description String Result Nothing data class BuyCryptoInfo val username String val phoneNumber PhoneNumber val creditCard CreditCard val kycVerification KycVerificationData with if null return "missing phone number" if null return "missing credit card" if null return "missing kyc verification" return Perfect now we handle the errors! Let’s try again to buy! Click… . Fill in the phone… Buy! . Fill in the credit card… Buy! … missing phone missing credit card missing kyc verification Let’s change the code again… data {

 companion object{
 fun from1(user: CryptoUser): Result<BuyCryptoInfo> = (user){
 val listOfErrors = mutableListOf< >() (phoneNumber == ) listOfErrors.add( ) (creditCard == ) listOfErrors.add( ) (kycVerification == ) listOfErrors.add( ) (listOfErrors.size > ) Result.Error(listOfErrors.joinToString( )) Result.Ok(BuyCryptoInfo(username, phoneNumber!!, creditCard!!, kycVerification!!))
 }
 }
} ( : , : , : , : ) class BuyCryptoInfo val username String val phoneNumber PhoneNumber val creditCard CreditCard val kycVerification KycVerificationData with String if null "missing phone number" if null "missing credit card" if null "missing kyc verification" if 0 return "," return Now the user gets all the errors at once. And he is satisfied… but , as a developer? Are you satisfied as well? what about you Konad to the rescue! I was not happy at all with solutions like the one above. Then I started investigating for a better way. That was a long trip. I went through and Monads, Applicative Functors, Higher-Kinded types. In the end, I realized that what I was searching for for Kotlin. We have , but I think that it is an overkill for this specific purpose. And also, it has a steep learning curve for an OOP developer. didn’t exist yet Arrow Then I decided to build my own solution and finally, I came out with . Konad With Konad, the methods above become as follows: io.konad.* ( username: String, phoneNumber: PhoneNumber, creditCard: CreditCard, kycVerification: KycVerificationData){ { : Result<BuyCryptoInfo> = with(user){
 ::BuyCryptoInfo.curry()
 .on(username)
 .on(phoneNumber.ifNull( ))
 .on(creditCard.ifNull( ))
 .on(kycVerification.ifNull( ))
 .result
 } : BuyCryptoInfo? = with(user){
 ::BuyCryptoInfo.curry()
 .on(username)
 .on(phoneNumber.maybe)
 .on(creditCard.maybe)
 .on(kycVerification.maybe)
 .nullable
 }
 }
} import data class BuyCryptoInfo val val val val companion object // You if you want to go for errors fun from1 (user: ) CryptoUser "missing phone number" "missing credit card" "missing kyc verification" // You if you want to go just for a null fun from2 (user: ) CryptoUser Hopefully, this can be considered pretty much cleaner than the previous proposals. Let’s finally see how to become rich! : = user = CryptoUser(username = , ..., ...) crypto = CryptoInfo( ) youGotRich1: Result< > = ::becomeRich.curry()
 .on(crypto)
 .on(BuyCryptoInfo.from1(user))
 .result youGotRich2: Result< > = BuyCryptoInfo.from1(user)
 .map { cryptoInfo -> becomeRich(crypto, cryptoInfo) } (youGotRich1 ){ Result.Ok -> Result.Errors -> youGotRich1.description
}.run(::println) fun becomeRich (crypto: , buyInfo: ) CryptoInfo BuyCryptoInfo Boolean false val "foo.bar" val "Bitcoin" val Boolean // or val Boolean when /*or youGotRich2*/ is "Congrats!" is // Will print Congrats! or the list of the missing information Conclusion Kotlin nullables are perfect to express the optionality of a computation. Anyway, they lack some UX in case of some more advanced use-cases. My hope is to have filled that gap with , and that you will enjoy using it. Konad Any feedback is much appreciated, as well as contributions to the library and bug or conceptual mistakes reporting. You can contact me on Twitter at @luca_picci or commenting the article. Thank you for reading.

Kotlin Nullables: How to Compose Them Right 

Recently nullable reference types have become trendy. Meanwhile, the good old nullable value types are still here and actively used. How well do you remember the nuances of working with them? Let's jog your memory or test your knowledge by reading this article. Examples of C# and IL code, references to the CLI specification, and CoreCLR code are provided. Let's start with an interesting case. . If you are interested in nullable reference types, you can read several articles by my colleagues: " ", " ". Note Nullable Reference types in C# 8.0 and static analysis Nullable Reference will not protect you, and here is the proof Take a look at the sample code below and answer what will be output to the console. And, just as importantly, why. Just let's agree right away that you will answer as it is: without compiler hints, documentation, reading literature, or anything like that. :) Well, let's do some thinking. Let's take a few main lines of thought that I think may arise. 1. Assume that int? is a reference type. Let's reason, that is a reference type. In this case, will be stored in , and it will also be stored in after assignment. A reference to an object will be stored in . It will also be stored in after assignment. As a result, will take and a non-null reference to the object as arguments, so... int? null a aObj b bObj Object.ReferenceEquals null That needs no saying, the answer is False! 2. Assume that int? is a value type. Or maybe you doubt that is a reference type? And you are sure of this, despite the expression? Well, let's go from the other side and start from the fact that is a value type. int? int? a = null int? In this case, the expression looks a bit strange, but let's assume that C# got some extra syntactic sugar. Turns out, stores an object. So does . When initializing and variables, objects stored in and will be boxed, resulting in different references being stored in and . So, in the end, takes references to different objects as arguments, therefore... int? a = null a b aObj bObj a b aObj bObj Object.ReferenceEquals That needs no saying, the answer is False! 3. We assume that here we use Nullable<T> . Let's say you didn't like the options above. Because you know perfectly well that there is no , but there is a value type , and in this case will be used. You also realize that and will actually have the same objects. With that, you remember that storing values in and will result in boxing. At long last, we'll get references to different objects. Since gets references to the different objects... int? Nullable<T> Nullable<int> a b aObj bObj Object.ReferenceEquals That needs no saying, the answer is False! 4. ;) For those who started from value types - if a suspicion crept into your mind about comparing links, you can view the documentation for at . In particular, it also touches on the topic of value types and boxing/unboxing. Except for the fact that it describes the case, when instances of value types are passed directly to the method, whereas we made the boxing separately, but the main point is the same. Object.ReferenceEquals docs.microsoft.com When comparing value types, if objA and objB are value types, they are boxed before they are passed to the ReferenceEquals method. This means that if both objA and objB represent the same instance of a value type , the ReferenceEquals method nevertheless returns false , as the following example shows. Here we could have ended the article, but the thing is that... the correct answer is . True Well, let's figure it out. Investigation There are two ways - simple and interesting. Simple way is . Open , where we look at the section "Boxing and Unboxing". Well, that's all, see the behavior description. But if you want more details, welcome to the interesting path. ;) int? Nullable<int> documentation on Nullable<T> Interesting way There won't be enough documentation on this path. It describes the behavior, but does not answer the question 'why'? What are actually and in the given context? Why does it work like this? Are there different commands used in the IL code or not? Is behavior different at the CLR level? Is it another kind of magic? int? null Let's start by analyzing the entity to recall the basics, and gradually get to the initial case analysis. Since C# is a rather "sugary" language, we will sometimes refer to the IL code to get to the bottom of things (yes, C# documentation is not our cup of tea today). int? int?, Nullable<T> Here we will look at the basics of nullable value types in general: what they are, what they are compiled into in IL, etc. The answer to the question from the case at the very beginning of the article is discussed in the next section. Let's look at the following code fragment: Although the initialization of these variables looks different in C#, the same IL code will be generated for all of them. As you can see, in C# everything is heartily flavored with syntactic sugar for our greater good. But in fact: is a value type. int? is the same as The IL code works with int? Nullable<int>. Nullable<int32> is the same as . In IL, this is compiled to an instruction that performs default initialization by the loaded address. int? aVal = null Nullable<int> aVal = new Nullable<int>() initobj Let's consider this code: We're done with the default initialization - we saw the related IL code above. What happens here when we want to initialize with the value 62? aVal Look at the IL code: Again, nothing complicated - the address pushes onto the evaluation stack, as well as the value 62. After the constructor with the signature is called. In other words, the following two statements will be completely identical: aVal Nullable<T>(T) You can also see this after checking out the IL code again: And what about the checks? What does this code represent? That's right, for better understanding, we will again refer to the corresponding IL code. As you may have guessed, there is actually no - all that happens is accessing the property. In other words, the same logic in C# can be written more explicitly in terms of the entities used, as follows. null Nullable<T>.HasValue IL code: Let's recap. Nullable value types are implemented using the type; Nullable<T> is actually a constructed type of the unbound generic value type ; int? Nullable<T> is the initialization of an object of type with the default value, no is actually present here; int? a = null Nullable<int> null - again, there is no , there is a call of the property. if (a == null) null Nullable<T>.HasValue The source code of the type can be viewed, for example, on GitHub in the dotnet/runtime repository - a . There's not much code there, so check it out just for kicks. From there, you can learn (or recall) the following facts. Nullable<T> direct link to the source code file For convenience, the type defines: Nullable<T> implicit conversion operator from to >; T Nullable<T explicit conversion operator from to . Nullable<T> T The main logic of work is implemented by two fields (and corresponding properties): - the value itself, the wrapper over which is ; T value Nullable<T> - the flag indicating "whether the wrapper contains a value". It's in quotation marks, since in fact always contains a value of type . bool hasValue Nullable<T> T Now that we've refreshed our memory about nullable value types, let's see what's going on with the boxing. Nullable<T> boxing Let me remind you that when boxing an object of a value type, a new object will be created on the heap. The following code snippet illustrates this behavior: The result of comparing references is expected to be . It is due to 2 boxing operations and creating of 2 objects whose references were stored in and false obj1 obj2 Now let's change to . int Nullable<int> The result is expectedly . false And now, instead of 62, we write the default value. Aaand... the result is unexpectedly . One might wonder that we have all the same 2 boxing operations, two created objects and references to two different objects, but the result is ! true true Yeah, it's probably sugar again, and something has changed at the IL code level! Let's see. Example N1. C# code: IL code: Example N2. C# code: IL code: Example N3. C# code: IL code: As we can see, in all cases boxing happens in the same way - values of local variables are pushed onto the evaluation stack ( instruction). After that the boxing itself occurs by calling the command, which specifies what type we will be boxing. ldloc box Next we refer to , see the description of the command, and find an interesting note regarding nullable types: Common Language Infrastructure specification box If typeTok is a value type, the box instruction converts val to its boxed form. ... If it is a nullable type, this is done by inspecting val's HasValue property; if it is false, a null reference is pushed onto the stack; otherwise, the result of boxing val's Value property is pushed onto the stack. This leads to several conclusions that dot the 'i': the state of the object is taken into account (the flag we discussed earlier is checked). If does not contain a value ( - ), the result of boxing is ; Nullable<T> HasValue Nullable<T> HasValue false null if contains a value ( - ), it is not a object that is boxed, but an instance of type that is stored in the field of type >; Nullable<T> HasValue true Nullable<T> T value Nullable<T specific logic for handling boxing is not implemented at the C# level or even at the IL level - it is implemented in the CLR. Nullable<T> Let's go back to the examples with that we touched upon above. Nullable<T> First: The state of the instance before the boxing: -> ; T int -> ; value 62 -> . hasValue true The value 62 is boxed twice. As we remember, in this case, instances of the type are boxed, not . Then 2 new objects are created, and 2 references to different objects are obtained, the result of their comparing is . int Nullable<int> false Second: The state of the instance before the boxing: -> ; T int -> (in this case, - a default value for ); value default 0 int -> . hasValue false Since is is , objects are not created. The boxing operation returns which is stored in variables and . Comparing these values is expected to return . hasValue false null obj1 obj2 true In the original example, which was at the very beginning of the article, exactly the same thing happens: For the sake of interest, let's look at the CoreCLR source code from the repository mentioned earlier. We are interested in the file , specifically, the x method with the logic we need: dotnet/runtime object.cpp Nullable::Bo Here we have everything we discussed earlier. If we don't store the value, we return : NULL Otherwise we initiate the boxing: Conclusion You're welcome to show the example from the beginning of the article to your colleagues and friends just for kicks. Will they give the correct answer and justify it? If not, share this article with them. If they do it - well, kudos to them! I hope it was a small but exciting adventure. :) Someone might have a question: how did we happen to dig that deep in this topic? We were writing a new diagnostic rule in related to working with arguments, one of which is represented by a value type. Suddenly it turned out that with there is an unexpected subtlety in the behavior when boxing. We looked at the IL code - there was nothing special about the . Checked out the CLI specification - and gotcha! The case promised to be rather exceptional and noteworthy, so here's the article right in front of you. P.S. PVS-Studio Object.ReferenceEquals Nullable<T> box By the way, recently, I have been spending more time on Twitter where I post some interesting code snippets and retweet some news in the .Net world and so on. Feel free to look through it and follow me if you want ( ). P.P.S. link to the profile Previously published at https://viva64.com/en/b/0772/

The Programmer's Exam: Nullable Value Types

C# 8.0 just rolled out with plenty of new features. One of the most important is the support of nullable reference types (NRT). A bunch of words that don't seem to explain what it does. I mean, aren't all types (except value types) already nullable? What Microsoft means by NRT is that now you can explicitly tell the compiler to check for null values. They do that by switching the default from null to not null for all types. So unless you explicitly specify it, you can assume that variables cannot be null anymore. Why do that? Well it's the first attempt by Microsoft to fix the dreaded NullReferenceException that has plagued so many applications. Many have even called the decision to allow null values in C# the . You can also look at my other post ( ) for my take on the subject. billion-dollar mistake Why null in C# is so bad Let's try it First of all, to try it, you must activate it. Either create a new application that runs on or open your and add this. .net core 3.0 .csproj 8.0 Enable < > LangVersion </ > LangVersion < > Nullable </ > Nullable is to enable the C# 8.0 features and is to switch the compiler behavior from all types by default to all types by default. <LangVersion>8.0</LangVersion> <Nullable>Enable</Nullable> null not null You can also enable/disable this feature only for a specific block of code using these instructions #nullable enable #nullable disable #nullable restore The compiler will then run static analysis on your code and add a compilation warning when it detects that a value can be null in a specific code path. e.g. a = ; l = a.Length; // warning CS8600: Converting null literal or possible null value to non-nullable type string null // warning CS8602: Dereference of a possibly null reference var New operators This feature also introduces two new operators to deal with those cases. `?` to tell the compiler that this object can be null. This operator has the same syntax as the one used by the nullable value type. e.g. int? `!` a.k.a. the null forgiving operator is used to tell the compiler that we know that the value won't be null. The compiler will not check anymore for this variable. Be careful, though, because if you are wrong and the value is null at runtime, you'll get the good old NullReferenceException. Anyways, this is still very useful, because even the mighty c# compiler can be wrong and get mistaken while analyzing your code. Those false positives tend to happen when you are using features like reflection or deserialization.Show me how to do it If we go back to the previous block of code, we now have all the tools to get rid of those warnings. (Note that in this case, those warnings are legit, and you should handle them in another way. It's just an example to demonstrate how to use those operators.) ? a = ; l = a.Length; // this will make sure we can assign a null value to that variable string null // this will remove the warning and let us try to call `.Length` var Conclusion NRT is a powerful tool that we'll all need to learn to use. It can bring us closer to with the elimination of some runtime problems by discovering them in the earliest stages of your Software Development Life Cycle (SDLC). This practice is also known as Shift Left. The idea is to raise issues sooner in the SDLC. A problem found during development costs way less that one found in production. correctness All code samples are available on github References https://docs.microsoft.com/en-us/dotnet/csharp/nullable-references Previously published at https://blog.miguelbernard.com/c-8-0-nullable-reference-types-are-here/

C# 8.0 Nullable Reference Types Are Here

Evolution of Data Sharding Towards Automation and Flexibility in Apache Doris

অ্যাপাচি ডরিসে অটোমেশন এবং নমনীয়তার দিকে ডেটা ভাগ করার বিবর্তন

Entwicklung des Data Sharding hin zu Automatisierung und Flexibilität in Apache Doris

Evolución de la fragmentación de datos hacia la automatización y la flexibilidad en Apache Doris

Evolution du partage de données vers l'automatisation et la flexibilité dans Apache Doris

अपाचे डोरिस में स्वचालन और लचीलेपन की ओर डेटा शार्डिंग का विकास

Apache Doris における自動化と柔軟性に向けたデータ シャーディングの進化

Apache Doris의 자동화 및 유연성을 향한 데이터 샤딩의 진화

Evolução da fragmentação de dados em direção à automação e flexibilidade no Apache Doris

Эволюция сегментирования данных в сторону автоматизации и гибкости в Apache Doris

Apache Doris'te Veri Parçalamanın Otomasyona ve Esnekliğe Doğru Evrimi

Sự phát triển của việc phân chia dữ liệu theo hướng tự động hóa và linh hoạt trong Apache Doris

Apache Doris 中数据分片向自动化和灵活性的演变

Authors:
(1) Gonzalo J. Aniano Porcile, LinkedIn;
(2) Jack Gindi, LinkedIn;
(3) Shivansh Mundra, LinkedIn;
(4) James R. Verbus, LinkedIn;
(5) Hany Farid, LinkedIn and University of California, Berkeley. Table of Links Abstract and Intro
Data sets
Model
Results
Discussion, Acknowledgements, and References 2. Data Sets Our training and evaluation leverage 18 data sets consisting of 120,000 real LinkedIn profile photos and 105,900 AI-generated faces spanning five different GAN and five different diffusion synthesis engines. The AI-generated images consist of two main categories, those with a face and those without. Real and synthesized color (RGB) images are resized from their original resolution to 512 × 512 pixels. Shown in Table 1 is an accounting of these images, and shown in Figure 2 are representative examples from each of the AI-generated categories as described next. 2.1. Real Faces The 120,000 real photos were sampled from LinkedIn users with publicly-accessible profile photos uploaded between January 1, 2019 and December 1, 2022. These accounts showed activity on the platform on at least 30 days (e.g., signed in, posted, messaged, searched) without triggering any fake-account detectors. Given the age and activity on the accounts, we can be confident that these photos are real. These images were of widely varying resolution and quality. Although most of these images are standard profile photos consisting of a single person, some do not contain a face. In contrast, all of the AI-generated images (described next) consist of a face. We will revisit this difference between real and fake images in Section 4. 2.2. GAN Faces 2, and 3, color images were synthesized at a resolution of 1024×1024 pixels and with ψ = 0.5. [1] For EG3D (Efficient Geometry-aware 3D Generative Adversarial Networks), the so-called 3D version of StyleGAN, we synthesized 10,000 images at a resolution of 512×512, with ψ = 0.5, and with random head poses. A total of 10,000 images at a resolution of 1024 × 1024 pixels were downloaded from generated.photos[2]. These GAN-synthesized images generally produce more professional looking head shots because the network is trained on a dataset of high-quality images recorded in a photographic studio. 2.3. GAN Non-Faces A total of 5,000 StyleGAN 1 images were downloaded[3] for each of three non-face categories: bedrooms, cars, and cats (the repositories for other StyleGAN versions do not provide images for categories other than faces). These images ranged in size from 512 × 384 (cars) to 256 × 256 (bedrooms and cats). 2.4. Diffusion Faces We generated 9,000 images from each Stable Diffusion [26] version (1, 2)[4]. Unlike the GAN faces described above, text-to-image diffusion synthesis affords more control over the appearance of the faces. To ensure diversity, 300 faces for each of 30 demographics with the prompts “a photo of a {young, middle-aged, older} {black, east-asian, hispanic, south-asian, white} {woman, man}.” These images were synthesized at a resolution of 512 × 512. This dataset was curated to remove obvious synthesis failures in which, for example, the face was not visible. An additional 900 images were synthesized from the most recent version of Stable Diffusion (xl). Using the same demographic categories as before, 30 images were generated for each of 30 categories, each at a resolution of 768 × 768. We generated 9,000 images from DALL-E 2 [5], consisting of 300 images for each of 30 demographic groups. These images were synthesized at a resolution of 512×512 pixels. A total of 1,000 Midjourney[6] images were downloaded at a resolution of 512 × 512. This images were manually curated to consist of only a single face. 2.5. Diffusion Non-Faces We synthesized 1,000 non-face images from each of two versions of Stable Diffusion (1, 2). These images were generated using random captions (generated by ChatGPT) and were manually reviewed to remove any images containing a person or face. These images were synthesized at a resolution of 600 × 600 pixels. A similar set of 1,000 DALL-E 2 and 1,000 Midjourney images were synthesized at a resolution of 512 × 512. 2.6. Training and Evaluation Data The above enumerated sets of images are split into training and evaluation as follows. Our model (described in Section 3) is trained on a random subset of 30,000 real faces and 30,000 AI-generated faces. The AI-generated faces are comprised of a random subset of 5,250 StyleGAN 1, 5,250 StyleGAN 2, 4,500 StyleGAN 3, 3,750 Stable Diffusion 1, 3,750 Stable Diffusion 2, and 7,500 DALL-E 2 images. We evaluate our model against the following: • A set of 5,000 face images from the same synthesis engines used in training (StyleGAN 1, StyleGAN 2, StyleGAN 3, Stable Diffusion 1, Stable Diffusion 2, and DALL-E 2). • A set of 5,000 face images from synthesis engines not used in training (Generated.photos, EG3D, Stable Diffusion xl, and Midjourney). • A set of 3,750 non-face images from each of five synthesis engines (StyleGAN 1, DALL-E 2, Stable Diffusion 1, Stable Diffusion 2, and Midjourney). • A set of 13,750 real faces. This paper is available on arxiv under CC 4.0 license. [1] The StyleGAN parameter ψ (typically in the range [0, 1]) controls the truncation of the seed values in the latent space representation used to generate an image. Smaller values of ψ provide better image quality but reduce facial variety. A mid-range value of ψ = 0.5 produces relatively artifact-free faces, while allowing for variation in the gender, age, and ethnicity in the synthesized face. [2] https://generated.photos/faces [3] https://github.com/NVlabs/stylegan) [4] https : / / github . com / Stability - AI / StableDiffusion [5] https://openai.com/dall-e-2 [6] https://www.midjourney.com Authors: (1) Gonzalo J. Aniano Porcile, LinkedIn; (2) Jack Gindi, LinkedIn; (3) Shivansh Mundra, LinkedIn; (4) James R. Verbus, LinkedIn; (5) Hany Farid, LinkedIn and University of California, Berkeley. Authors: Authors: (1) Gonzalo J. Aniano Porcile, LinkedIn; (2) Jack Gindi, LinkedIn; (3) Shivansh Mundra, LinkedIn; (4) James R. Verbus, LinkedIn; (5) Hany Farid, LinkedIn and University of California, Berkeley. Table of Links Abstract and Intro Data sets Model Results Discussion, Acknowledgements, and References Abstract and Intro Abstract and Intro Data sets Data sets Model Model Results Results Discussion, Acknowledgements, and References Discussion, Acknowledgements, and References 2. Data Sets Our training and evaluation leverage 18 data sets consisting of 120,000 real LinkedIn profile photos and 105,900 AI-generated faces spanning five different GAN and five different diffusion synthesis engines. The AI-generated images consist of two main categories, those with a face and those without. Real and synthesized color (RGB) images are resized from their original resolution to 512 × 512 pixels. Shown in Table 1 is an accounting of these images, and shown in Figure 2 are representative examples from each of the AI-generated categories as described next. 2.1. Real Faces The 120,000 real photos were sampled from LinkedIn users with publicly-accessible profile photos uploaded between January 1, 2019 and December 1, 2022. These accounts showed activity on the platform on at least 30 days (e.g., signed in, posted, messaged, searched) without triggering any fake-account detectors. Given the age and activity on the accounts, we can be confident that these photos are real. These images were of widely varying resolution and quality. Although most of these images are standard profile photos consisting of a single person, some do not contain a face. In contrast, all of the AI-generated images (described next) consist of a face. We will revisit this difference between real and fake images in Section 4. 2.2. GAN Faces 2, and 3, color images were synthesized at a resolution of 1024×1024 pixels and with ψ = 0.5. [1] For EG3D (Efficient Geometry-aware 3D Generative Adversarial Networks), the so-called 3D version of StyleGAN, we synthesized 10,000 images at a resolution of 512×512, with ψ = 0.5, and with random head poses. A total of 10,000 images at a resolution of 1024 × 1024 pixels were downloaded from generated.photos[2]. These GAN-synthesized images generally produce more professional looking head shots because the network is trained on a dataset of high-quality images recorded in a photographic studio. 2.3. GAN Non-Faces A total of 5,000 StyleGAN 1 images were downloaded[3] for each of three non-face categories: bedrooms, cars, and cats (the repositories for other StyleGAN versions do not provide images for categories other than faces). These images ranged in size from 512 × 384 (cars) to 256 × 256 (bedrooms and cats). 2.4. Diffusion Faces We generated 9,000 images from each Stable Diffusion [26] version (1, 2)[4]. Unlike the GAN faces described above, text-to-image diffusion synthesis affords more control over the appearance of the faces. To ensure diversity, 300 faces for each of 30 demographics with the prompts “a photo of a {young, middle-aged, older} {black, east-asian, hispanic, south-asian, white} {woman, man}.” These images were synthesized at a resolution of 512 × 512. This dataset was curated to remove obvious synthesis failures in which, for example, the face was not visible. An additional 900 images were synthesized from the most recent version of Stable Diffusion (xl). Using the same demographic categories as before, 30 images were generated for each of 30 categories, each at a resolution of 768 × 768. We generated 9,000 images from DALL-E 2 [5], consisting of 300 images for each of 30 demographic groups. These images were synthesized at a resolution of 512×512 pixels. A total of 1,000 Midjourney[6] images were downloaded at a resolution of 512 × 512. This images were manually curated to consist of only a single face. 2.5. Diffusion Non-Faces We synthesized 1,000 non-face images from each of two versions of Stable Diffusion (1, 2). These images were generated using random captions (generated by ChatGPT) and were manually reviewed to remove any images containing a person or face. These images were synthesized at a resolution of 600 × 600 pixels. A similar set of 1,000 DALL-E 2 and 1,000 Midjourney images were synthesized at a resolution of 512 × 512. 2.6. Training and Evaluation Data The above enumerated sets of images are split into training and evaluation as follows. Our model (described in Section 3) is trained on a random subset of 30,000 real faces and 30,000 AI-generated faces. The AI-generated faces are comprised of a random subset of 5,250 StyleGAN 1, 5,250 StyleGAN 2, 4,500 StyleGAN 3, 3,750 Stable Diffusion 1, 3,750 Stable Diffusion 2, and 7,500 DALL-E 2 images. We evaluate our model against the following: • A set of 5,000 face images from the same synthesis engines used in training (StyleGAN 1, StyleGAN 2, StyleGAN 3, Stable Diffusion 1, Stable Diffusion 2, and DALL-E 2). • A set of 5,000 face images from synthesis engines not used in training (Generated.photos, EG3D, Stable Diffusion xl, and Midjourney). • A set of 3,750 non-face images from each of five synthesis engines (StyleGAN 1, DALL-E 2, Stable Diffusion 1, Stable Diffusion 2, and Midjourney). • A set of 13,750 real faces. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv [1] The StyleGAN parameter ψ (typically in the range [0, 1]) controls the truncation of the seed values in the latent space representation used to generate an image. Smaller values of ψ provide better image quality but reduce facial variety. A mid-range value of ψ = 0.5 produces relatively artifact-free faces, while allowing for variation in the gender, age, and ethnicity in the synthesized face. [2] https://generated.photos/faces [3] https://github.com/NVlabs/stylegan) [4] https : / / github . com / Stability - AI / StableDiffusion [5] https://openai.com/dall-e-2 [6] https://www.midjourney.com

NULAB

nulab-inc.com

-- employees

Since n.d.

COMPANY RANKING

NULAB

EVERGREEN INDEX #7513

Representing this company? Click here to claim and customize this page!

Nulab WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!

NULAB

nulab-inc.com

-- employees

Since n.d.

COMPANY RANKING

NULAB

EVERGREEN INDEX #7513

RELATED COMPANIESRANK

STORIES ABOUT NULAB

Representing this company? Click here to claim and customize this page!

Nulab WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!