How Alexey Moiseenkov reached Prisma and went further
Prisma - an application for processing photos in the style of certain artists - has become one of the examples of sharp popularity in the global market. At the same time, Prism received both a mass of admiring praises and a number of reproaches: from the secondary development to the short-term popularity of the product itself. From the beginning and to this day, Alexey Moiseenkov is the main engine of this project. Next is our conversation with him.
The hype has passed. Did you use it correctly? Maybe there are circumstances that did not allow you to use the hyip sufficiently?
Alexey Moiseenkov: There is no secret how to use HYIP to the maximum, but we have benefited sufficiently. Got enough experience. Our use of HYIP was not in monetization, but in the experience of rapid growth around the world. One of the main things we received was the experience of rapid development and the experience of prompt communication with the press.
')
Prisma would never be in demand for long. Photo styling is like fashion. Neither we, no one can do anything to make the hype last forever. Of course, we could have extracted more money and connections; we could, of course, come up with a lot more methods of working with the audience. It would be worth working more actively in Asia. But this is not known for sure; it is not clear how these steps would have affected the popularity of Prism.
You could think of something else on top of the Prism. But the rush to artistic processing of the photo was asleep, and further the stylization will live within other products, such as “masks”.
In a matter of months, we had to tackle simultaneously the problem of a sharp increase in the company's staff, an increase in the productivity of the application, to interact with investors, so in a sense I was not in a bit of a search for a new hyip.
If such a growth fell to the share of a large company, it would probably launch more product options. On the product steps that we did and which could be done, you need to write a separate large article.
Well, what have you learned this year? You know, before Prism, the MIPT students were taught on the “How to make your startup” course. What changes would you make now in your course and in terms of the success of startups?
Alexey: I did not have one recipe for success. No, and now. I accumulated in that course my experience and thoughts from the books of other business authors. Now I would make changes, but changes need to be made throughout the course. I would not have distinguished one thing in the success of a startup before, and, in particular, I would not be able to use 2016-2017 as well. I can find a course and post it if in the comments to the interview someone asks him.
You are done with programming because of boredom. You like to generate ideas and see their realization. Now, instead of new ideas, a business routine has gone (hiring people, working corporately, thinking through a strategy) - does this not prevent us from maintaining an interest in the work?
Alexey: I do not see routine in business routine. I liked programming. I liked the feeling that something done by hand was working. But tired. I know that I can code, I can sell, and it's not so interesting to repeat. In business, I think, I still do not have to be bored and never have to. Here, unlike programming, everything changes quickly. In the morning some worries, in the evening others.
You have had experience in the past and abandoned projects, and low demand in the app on the running application. How can you keep motivation in such situations in order to finish the product and overcome low demand?
Alexey: Only the leader and his character will help here. Motivation is always kept on the leader. Not always success, but always the motivation of the team. I did not program Prism at all. I realized that the idea of ​​Prism would fly, and convinced my team to do it. The leader must believe in the idea. It is easy to understand who the leader is - who is being interviewed; that is usually what motivates within companies. Here we give an interview. By the way, Mapoji - the last start-up of our team (though not yet without Ilya) - was slowed down not by low download rates, but it was just a complicated project. We did not pull this unwind. Any social projects and networks are heavy on the rise. Complex and any content project.
You mentioned Snapchat. I already wrote that Evan Spiegel skillfully goes from HYIP to HYIP (disappearing cribs, sex provocations, masks, day reports), rolling out a new chip for his messenger every year, and Zuckerberg without any new chips is growing on the audience and advertising. Which way is closer to you?
Alexey: The sign of the HYIP project is not the final popularity, but the fact that everyone remembers at once how it was growing rapidly. I think that, for example, and Snapchat did not detonate the market immediately. Who remembers how many installations he had in the early years? Only experts. And all the HYIP projects are remembered by themselves.
Now we are making a product for regular use. So, rather, the way Zuckerberg. But by the way, I disagree with the thesis that it was possible to do, like Snapchat, and go from HYIP to HYIP. First, as I said, we did not have time. Secondly, Snapchat is not a hyip. Disappearing reportage news (“stories,” as they are called on Snapchat) is a daily need. The nonsense that you share with close friends is no longer interesting to anyone. And do not show it to anyone else. It's too much. It is important only to your friends. Yes, at first they had disappearing pictures, not videos, but the current “stories” are their logical continuation. But in my opinion, the essence remains the same. Disappearing news is relevant here and now. You do not load the memory and do not force to rake informational debris later. In a sense, this is a way to get rid of information overload.
How do you organize your work and team work? Slack, it is clear that you are using, but is there a place for a methodology like Agile or some kind of your own?
Alexey: Methodology is an ideal. All consultants write the ideal, and any company simply strives for it, but never achieves it. We work on the system of chaos. And do not spend time on procedures. There is a backlog, because it is convenient, but we have no project manager and a separate project manager either. Planning in Trello. And we realize everything to whom it is convenient. It is important for me that there is transparency in understanding who does what. Here, my experience in large companies affects. There is often a course of the project is not clear. In a team of 25 people to see the picture is not so difficult. This transparency of the process is provided by Trello.
How competent are you in recognizers and in neural networks? Did you write the neural network yourself? How well did she recognize?
Alexey: Personally, I myself wrote and wrote neural networks, but not very complicated ones. On the basis of libraries TensorFlow I write. I have a diploma from Polytech SPb in ranking in the search, a second diploma in regression models, that is, in classical machine learning. I, maybe, by recognizers, I cannot act as a technical expert or get a job, but I wrote. Recognition errors were at 20% for the top 1, for the top 5 there were 10% errors. My last neural network was written a year and a half ago.
What is the fundamental difference in the methodology for processing static images and dynamic video? In your programs, the optimization goes through the speed of iteration when repeating operations or is there a nontrivial algorithm?
Alexey: The standard approach is to perceive the video as a set of frames, that is, just a lot of pictures flashing in a row. You can also use the analysis of the relationship between frames. This is how the algorithms of any compression or streaming video work. When analyzing relationships, you can not produce unnecessary actions to process each frame. But most neural networks take exactly every frame into processing. We have the same way. No magic, no algorithms for analyzing the relationships of the frame between each other.
Do you agree with the thesis that we do not fully understand how a neural network works?
Alexey: Speech, rather, not about understanding, but about end-to-end tracking of all mathematical transformations that give the result at the output. We cannot connect all the paths of the transformation of the signs (in a particular layer or “neuron”) with the final result. But this is not a lack of understanding. We simply cannot physically analyze each transformation and describe the whole process. It is because of the lack of resources for the description of such a number of operations that we “do not understand everything”. We miss the way of converting weights, tensors and matrices for purely technical reasons, and not because of some kind of magic. True, most neural network architectures are made just on a whim. Trial and error. But this is one and the same reason - to try faster than to analyze and understand.
Does your new program PrismaLabs.ai have NIST rating and certification? They are from the Russian companies NtechLab or 3DiVi. Is it possible to compare your algorithms through this rating?
Alexey: I do not understand why I need to be rated. I think this is only necessary for working with the public sector. We are not working with the public sector yet, just no one has come from there yet. We are doing well without government customers. Therefore, we do not compare ourselves with others.
The scope of computer vision in the B2B sector, you called the identification of buyers in stores. Your other example is autopilot recognition of road signs. Where else can your application be used?
Alexey: Our goal is not the final functions, but the most effective neural network algorithms for porting computer vision to work specifically on mobile devices. And we are not competing with application solutions, but with a minimum of server resources spent on operation. I will explain. Here we can, for example, give users the effect of a " green screen " for changing the background when streaming games. But the stream usually comes from the desktop, not from the mobile, and there are better solutions. We specialize in an audience with mobile products. And therefore, substituting the background in streaming is pointless to promote.
How soon can a recognizer program detect a flying cruise missile? Or when will be able to follow the safety technique in the shop?
Aleksey: I was engaged in air defense and S-300 at the military department. But there were no such tasks, everything was solved through radar. But then there were no such technologies. I think the possibilities are already there, there is no mass application. In the short term, problems are solved. These are no longer creative tasks, but purely technical ones.
What is necessary for our programmers to stay in the country? You stayed here, because not a programmer, but already, rather, an entrepreneur?
Alexey: It's not about the level of payment. This is a broader question why people leave in the professional field. There are more interesting projects in the world than ours. Many friends want their project known in the world. From Russia such projects can hardly grow and take over the world. In America, the project will become popular faster. Many leave just for new sensations due to personal fatigue. In another place they find either personal or household comfort and get used to it.
Alexei himself has not changed much so far. Alexey lives in the same rented apartment as before. Just as before, he continues to play computer games. Three hours a day playing in Hearthstone. His BattleTag is darkolorin # 2275.
Alexey: Any interview is a distortion. No one is interested in the truth, any media adapts to the reader and fills the story with fiction. The story that, for example, two dudes were sitting and doing a startup at computers and all they needed was computers that would never be popular. Their program may be popular, but the media will always come up with something about its creation.
Business lessons
The hype has passed. Did you use it correctly? Maybe there are circumstances that did not allow you to use the hyip sufficiently?
Alexey Moiseenkov: There is no secret how to use HYIP to the maximum, but we have benefited sufficiently. Got enough experience. Our use of HYIP was not in monetization, but in the experience of rapid growth around the world. One of the main things we received was the experience of rapid development and the experience of prompt communication with the press.
')
Prisma would never be in demand for long. Photo styling is like fashion. Neither we, no one can do anything to make the hype last forever. Of course, we could have extracted more money and connections; we could, of course, come up with a lot more methods of working with the audience. It would be worth working more actively in Asia. But this is not known for sure; it is not clear how these steps would have affected the popularity of Prism.
The impetus for the creation of "Prism" was the graduation project of my student at the Moscow Institute of Physics and Technology, Ilya Frolov, who did the thing in the Style Transfer style. There were similar solutions already working. The task was to implement on the basis of someone else's idea their own processing model and their processing speeds, as well as to build up a base of style patterns. That is, the algorithm is not open source, but is based on an idea that lies in open source and then accelerated.
You could think of something else on top of the Prism. But the rush to artistic processing of the photo was asleep, and further the stylization will live within other products, such as “masks”.
In a matter of months, we had to tackle simultaneously the problem of a sharp increase in the company's staff, an increase in the productivity of the application, to interact with investors, so in a sense I was not in a bit of a search for a new hyip.
If such a growth fell to the share of a large company, it would probably launch more product options. On the product steps that we did and which could be done, you need to write a separate large article.
Facts from the biography of Alexei
- I started programming at 11 years old.
- In grade 11, he was in the top 200 Russian players in World of Warcraft.
- I earned my first money in the 8th grade - I painted with a stencil of T-shirts at the youth labor exchange.
- The first money from programming earned at the second year of the institute - he wrote and sold the program for 10,000 rubles.
- I always studied excellent and did not skip a single lesson.
- Prior to Prism, Alexey had already had experience selling a popular product in the Google Play app store — hundreds of thousands of downloads of a widget that allowed him to put a smiley and a signature on his desktop.
- Prizma initially had 4 co-founders.
- In order for Prism to start generating interest, it had to be transferred to mobile phones and accelerated a thousand times compared to the original desktop version, achieving processing speed on servers of 10 seconds.
- Advertising in TJournal gave an initial jump in 100 thousand downloads of Prism, and advertising in American Product Hunt - 5 million downloads.
- Now Prisma Labs is releasing tools for porting visual effects and computer vision to mobile devices, and also provides an opportunity to test them for its audience. The task is to earn on the sale of these services at $ 1 million per month.
- Prisma Labs' active user base is 5-10 million, mostly American users.
- Prisma Labs now specializes in transferring styles (starting product), recognizing and separating the background or scene, recognizing and highlighting an object (segmentation), masks for face photos, and recognizing emotions.
Well, what have you learned this year? You know, before Prism, the MIPT students were taught on the “How to make your startup” course. What changes would you make now in your course and in terms of the success of startups?
Alexey: I did not have one recipe for success. No, and now. I accumulated in that course my experience and thoughts from the books of other business authors. Now I would make changes, but changes need to be made throughout the course. I would not have distinguished one thing in the success of a startup before, and, in particular, I would not be able to use 2016-2017 as well. I can find a course and post it if in the comments to the interview someone asks him.
Recently, my former students told me that they didn’t trust me then - I didn’t have my own startup. And now they are listening to these lectures!
You are done with programming because of boredom. You like to generate ideas and see their realization. Now, instead of new ideas, a business routine has gone (hiring people, working corporately, thinking through a strategy) - does this not prevent us from maintaining an interest in the work?
Alexey: I do not see routine in business routine. I liked programming. I liked the feeling that something done by hand was working. But tired. I know that I can code, I can sell, and it's not so interesting to repeat. In business, I think, I still do not have to be bored and never have to. Here, unlike programming, everything changes quickly. In the morning some worries, in the evening others.
You have had experience in the past and abandoned projects, and low demand in the app on the running application. How can you keep motivation in such situations in order to finish the product and overcome low demand?
Alexey: Only the leader and his character will help here. Motivation is always kept on the leader. Not always success, but always the motivation of the team. I did not program Prism at all. I realized that the idea of ​​Prism would fly, and convinced my team to do it. The leader must believe in the idea. It is easy to understand who the leader is - who is being interviewed; that is usually what motivates within companies. Here we give an interview. By the way, Mapoji - the last start-up of our team (though not yet without Ilya) - was slowed down not by low download rates, but it was just a complicated project. We did not pull this unwind. Any social projects and networks are heavy on the rise. Complex and any content project.
You mentioned Snapchat. I already wrote that Evan Spiegel skillfully goes from HYIP to HYIP (disappearing cribs, sex provocations, masks, day reports), rolling out a new chip for his messenger every year, and Zuckerberg without any new chips is growing on the audience and advertising. Which way is closer to you?
Alexey: The sign of the HYIP project is not the final popularity, but the fact that everyone remembers at once how it was growing rapidly. I think that, for example, and Snapchat did not detonate the market immediately. Who remembers how many installations he had in the early years? Only experts. And all the HYIP projects are remembered by themselves.
Now we are making a product for regular use. So, rather, the way Zuckerberg. But by the way, I disagree with the thesis that it was possible to do, like Snapchat, and go from HYIP to HYIP. First, as I said, we did not have time. Secondly, Snapchat is not a hyip. Disappearing reportage news (“stories,” as they are called on Snapchat) is a daily need. The nonsense that you share with close friends is no longer interesting to anyone. And do not show it to anyone else. It's too much. It is important only to your friends. Yes, at first they had disappearing pictures, not videos, but the current “stories” are their logical continuation. But in my opinion, the essence remains the same. Disappearing news is relevant here and now. You do not load the memory and do not force to rake informational debris later. In a sense, this is a way to get rid of information overload.
Nobody knows how to keep HYIP. Most likely, this is impossible. We had two values ​​- technology and audience. The technology has already been copied. And there is an audience that is a shaky thing. Therefore, the goal of the technology company we created in the wake of the HYIP is to produce products that we use regularly.
How do you organize your work and team work? Slack, it is clear that you are using, but is there a place for a methodology like Agile or some kind of your own?
Alexey: Methodology is an ideal. All consultants write the ideal, and any company simply strives for it, but never achieves it. We work on the system of chaos. And do not spend time on procedures. There is a backlog, because it is convenient, but we have no project manager and a separate project manager either. Planning in Trello. And we realize everything to whom it is convenient. It is important for me that there is transparency in understanding who does what. Here, my experience in large companies affects. There is often a course of the project is not clear. In a team of 25 people to see the picture is not so difficult. This transparency of the process is provided by Trello.
Neural networks
How competent are you in recognizers and in neural networks? Did you write the neural network yourself? How well did she recognize?
Alexey: Personally, I myself wrote and wrote neural networks, but not very complicated ones. On the basis of libraries TensorFlow I write. I have a diploma from Polytech SPb in ranking in the search, a second diploma in regression models, that is, in classical machine learning. I, maybe, by recognizers, I cannot act as a technical expert or get a job, but I wrote. Recognition errors were at 20% for the top 1, for the top 5 there were 10% errors. My last neural network was written a year and a half ago.
What is the fundamental difference in the methodology for processing static images and dynamic video? In your programs, the optimization goes through the speed of iteration when repeating operations or is there a nontrivial algorithm?
Alexey: The standard approach is to perceive the video as a set of frames, that is, just a lot of pictures flashing in a row. You can also use the analysis of the relationship between frames. This is how the algorithms of any compression or streaming video work. When analyzing relationships, you can not produce unnecessary actions to process each frame. But most neural networks take exactly every frame into processing. We have the same way. No magic, no algorithms for analyzing the relationships of the frame between each other.
Do you agree with the thesis that we do not fully understand how a neural network works?
Alexey: Speech, rather, not about understanding, but about end-to-end tracking of all mathematical transformations that give the result at the output. We cannot connect all the paths of the transformation of the signs (in a particular layer or “neuron”) with the final result. But this is not a lack of understanding. We simply cannot physically analyze each transformation and describe the whole process. It is because of the lack of resources for the description of such a number of operations that we “do not understand everything”. We miss the way of converting weights, tensors and matrices for purely technical reasons, and not because of some kind of magic. True, most neural network architectures are made just on a whim. Trial and error. But this is one and the same reason - to try faster than to analyze and understand.
Does your new program PrismaLabs.ai have NIST rating and certification? They are from the Russian companies NtechLab or 3DiVi. Is it possible to compare your algorithms through this rating?
Alexey: I do not understand why I need to be rated. I think this is only necessary for working with the public sector. We are not working with the public sector yet, just no one has come from there yet. We are doing well without government customers. Therefore, we do not compare ourselves with others.
The scope of computer vision in the B2B sector, you called the identification of buyers in stores. Your other example is autopilot recognition of road signs. Where else can your application be used?
Alexey: Our goal is not the final functions, but the most effective neural network algorithms for porting computer vision to work specifically on mobile devices. And we are not competing with application solutions, but with a minimum of server resources spent on operation. I will explain. Here we can, for example, give users the effect of a " green screen " for changing the background when streaming games. But the stream usually comes from the desktop, not from the mobile, and there are better solutions. We specialize in an audience with mobile products. And therefore, substituting the background in streaming is pointless to promote.
Until July 2017, our application also worked on servers, and the task was to transfer neural networks technology to mobile phones. Everyone said it was unreal. And we managed. With our application, we have given birth to the neural network industry for mobile phones.
How soon can a recognizer program detect a flying cruise missile? Or when will be able to follow the safety technique in the shop?
Aleksey: I was engaged in air defense and S-300 at the military department. But there were no such tasks, everything was solved through radar. But then there were no such technologies. I think the possibilities are already there, there is no mass application. In the short term, problems are solved. These are no longer creative tasks, but purely technical ones.
What is necessary for our programmers to stay in the country? You stayed here, because not a programmer, but already, rather, an entrepreneur?
Alexey: It's not about the level of payment. This is a broader question why people leave in the professional field. There are more interesting projects in the world than ours. Many friends want their project known in the world. From Russia such projects can hardly grow and take over the world. In America, the project will become popular faster. Many leave just for new sensations due to personal fatigue. In another place they find either personal or household comfort and get used to it.
Alexei himself has not changed much so far. Alexey lives in the same rented apartment as before. Just as before, he continues to play computer games. Three hours a day playing in Hearthstone. His BattleTag is darkolorin # 2275.
P.S
Alexey: Any interview is a distortion. No one is interested in the truth, any media adapts to the reader and fills the story with fiction. The story that, for example, two dudes were sitting and doing a startup at computers and all they needed was computers that would never be popular. Their program may be popular, but the media will always come up with something about its creation.
Source: https://habr.com/ru/post/338758/
All Articles