HNHacker News
TopNewBestAskShowJobs

somethingsome

462 karma · joined June 13, 2020

Contact me at somethingsomehn@protonmail.com
submissionscomments
somethingsome··on Ask HN: I'm building a Windows video editor, what are your pain points?
I do research in imaging, so comparing images generated by algorithms consumes a lot of my time!

Side-by-side let me improve my understanding of the algorithms' trade-offs (think for example comparing different compression algorithms, inspecting artifacts). It would also help me debug my codes by comparing pixels values with a reference for example, and many other usages.

Of course, those are not the only requirements, but it would already spare me a lot of time to be able to open this kind of format by simply clicking on the file :)

The others operations I perform daily are frames' differences (|img1 - img2|) and display them using some colormap. We also frequently need to run several quality metrics like MS-PSNR, SSIM, etc.. to "objectively" evaluate what we see.

For the OpenEXR their lib is ok for opening images, I need this because several images I use have only one channel in 16bits, and it's one of the only formats capable of storing this kind of data.

While for YUV, the format is actually extremely simple! It's just a big array containing an intensity component (luma) and colors components (chroma) with some upscaling[1] and GPU's decoders can work with them if it's a performance issue [2] ;).

Both are standard in the professional video community! Think of them as "raw" video formats.

[1] https://stackoverflow.com/questions/60729170/python-opencv-c...

[2] https://docs.nvidia.com/drive/drive_os_5.1.6.1L/nvvib_docs/i...

somethingsome··on Ask HN: I'm building a Windows video editor, what are your pain points?
(semi pro - pro)

In every editor I used, they usually miss YUV support and OpenEXR support (monochannel 16bit images for example). It would also be terribly useful to me to be able to zoom anywhere in the frames and get the precise color values and location of any pixel.

A fast way to open two frames of the same file or of two different files side by side to compare would also be extremely useful!

My current work flow is to use

1.YUVView which cover YUV and zooming but it has different bugs [1]

2. OpenCV for OpenEXR, and use the debugger to inspect colors.. It is terribly painful to compare two images :)

If someone has better way to analyse this kind of files I'm very interested!

[1] https://github.com/IENT/YUView

somethingsome··on Ask HN: Is PhD the only way to be able to enter research?
> Is it?

DL/RL/etc. Are very hype right now in research, everything is moving at a very fast speed. A lot of 'RL applied to robotics' papers are accepted at top conferences (and even more rejected)

Several tools are in development to increase dramatically the speed at which research can be done ;)

A simple Google scholar search on reinforcement learning and robotics limited to 2020 only should give you lots of good results.

I wish you the best ;)

somethingsome··on Ask HN: Is PhD the only way to be able to enter research?
RL is extensively used in robotics ;)

If you want to go for a PhD, I suggest reading Sutton's book on RL (at least the important parts for what you want to do), followed by recent RL papers and RL papers in robotics that pick your interest, understand them fully. Aim for journal papers and major conferences only (high impact factor is a first metric that comes to mind).

The main bit is to try to formulate an important research question you're interested in which is not already explored in the vast litterature and plan which experiments you need to carry and what do you need to change if some of them fail.

If you want to engage in a PhD, an important and well studied research question will increase your chances of getting accepted in a PhD program.

The main advantage in doing a PhD in a uni is that you can share and learn knowledge from other students and advisors, as well as have insights in what are ideas worth pursuing and what are small improvement that no one care about.

At uni you usually work on the kind of research meant to improve the knowledge and less on the general engineering problems. Engineering problems are byproduct of the knowledge you are building. A lot of uni work very closely with industries in practical fields like robotics, you'll have plenty of opportunities to apply what you are building directly on engineering problems. But that will not be the main focus.

In my experience it is considered less 'valuable research' to try to apply some SoTA to a specific engineering problem in academy, and usually proposals of this kind are rejected.

Hope this is helpful :)

somethingsome··on Ask HN: Is PhD the only way to be able to enter research?
A PhD is usually where you learn what research _is_ and how it is done, that's the reason why it is usually a 'minimum requirement'

If you plan to spend some years in reading and implementing papers to become a researcher anyway.. You can probably go back to school and get a PhD, this will be way more efficient than randomly reading without guidance.

Furthermore, a researcher do not spend his days to implement state of the art algorithms, he tries to push science a little further in very small steps, implementing is just part of the process.

Even when you have a PhD, pure research jobs are pretty rare but from your description you are more looking into innovative startups that implement SoTA papers (more R&D than research), this is more common and most of the jobs in robotic require some experimental setup and "research". In those startup the PhD requirement is often flexible !

But this "research" is often very different from the research done in academy.

Try to clarify what kind of research you want to do so that people can give you more precise advices :)

somethingsome··on Eulerian Video Magnification
You are free to contact him! I'm not sure he is not willing to share.

Sharing code to the public in research depends on several factors, it can get open sourced when the thesis is done, it may depends on who is paying the research (eg: some industrial PhD are not public), long term plan exist to publish the code but there are only 24h a day and a whole lot of work to do, the author/uni may want to make a business after the thesis, and a lot of other parametres. It gets even worse when the code is used in medical applications as a validation process with all the relevant authorities has to be done. This can take a lot of time and freeze the code in the waiting for a certification. Also sharing code between researchers is very usual, it doesn't mean the code has to go public. In this case I think it is part because the original matlab code is patented, research is ongoing and as it is research code it is not necessarily public ready.

It doesn't mean he will never release it.

If you desperately need it, the matlab code is available, and has the same functionality, you are free to adapt it or use it directly (for research purposes).

I must say I don't like that the original code is patented, but I see hundreds of reasons why the authors patented it even if I don't particularly agree :)

somethingsome··on Eulerian Video Magnification
A friend reimplementated it in python and it worked pretty well for analyzing cardiac liver movements[1]!

I don't think he is willing to share the code, but my guess is that a python implementation should work just fine, with very nice results :) (I saw the results applied in his research on videos with impressive results). It took him something like several months to have a code without major bugs.

[1] Hahn, S., Absil, J., Debeir, O., & Metens, T. (2019). Assessment of cardiac‐driven liver movements with filtered harmonic phase image representation, optical flow quantification, and motion amplification. Magnetic resonance in medicine, 81(4), 2788-2798.

somethingsome··on Sets and Probability
(From the name of OP I deduce he may read this.)

You should read the first and second chapter of 'Probability Theory: the Logic of Science' by Jaynes. It offers a derivation of probability theory using only Logic and Boolean algebra as foundations. I found it delightful, I hope you will also enjoy it :)

It's somewhat more intensive than Taleb, but the derivations are really beautiful.

somethingsome··on Unlearn rotation matrices as rotations
Halmos are always good books. I didn't know he had this viewpoint on determinants though.

What I meant is that on a pure mathematical point of view they are seen as inelegant as they require a basis to be defined. I don't have a problem with that when you are doing maths on manifolds etc..

From a practical perspective, a lot of engineering problems go from 1 to 4 dimensions and are basis dependents. And the determinant becomes a useful tool.

somethingsome··on Unlearn rotation matrices as rotations
You went through Axler right?

It's a respectable idea that determinants should not be as important as they are on a pedagogical point of view imo. But they can be extremely useful.

From my point of view, matrices are a bad abstraction in general. For example if we have a linear or a bilinear form, both can be written as matrices of the same size, we can no more differentiate the two objects.

The trouble comes when we want to apply some transformation over those matrices.. as they do not use the same transformations laws.. Namely A^-1.L.A and A^T.B.A

This is a very usual error in computer graphics when we want to make some transformation on a 4x4 matrix and we do not know which kind of object it is.

somethingsome··on Coca-Cola paid scientists to downplay how sweet drinks fueled the obesity crisis
Very difficult, Sadly nutrition research is not yet very conclusive on what is good and what is bad for our health. (It's very difficult to have correct experiments as the volunteers never have the same physiology) And our nutriments needs may be very specific to one's genetics and activity level. Sugar is fuel for the brain and the body, it is not necessary bad, but abusing it is bad. Maybe sugar (eg pasta) is bad when mixed with other ingredients and healthy with others. It is extremely difficult to reach a good consensus even if you read the litterature.

And usually (micro)nutriments have trade-offs : good for somethings bad for others.

Look at the different kind of diet, they are all based on different researches and can be extremely different while all are using 'valid' research data. The main idea behind them is usually to have 'good' proportions between macronutrients (each diet define a different proportion) while keeping the number of kcal in the range you want to reach your objectives.

For a first rule of thumb that I would use: abusing in any direction is bad. Sugar is ok in 'small' quantities, fat, etc also. However reality is way more messy :)

somethingsome··on Ask HN: Is it possible to get PhD without completing bachelors
I would strongly advice against it.

It seems like you are looking at a PhD like "the diploma to have" but not realizing what a PhD is in practice. Read a lot of what a PhD consist in: At least, expect to be lonely and work a whole lot of hours by yourself without much guidance or external motivation, you are becoming the expert during a PhD, so you will not find a lot of people that can help you when you have a problem. (I don't say that good and knowledgeable advisors do not exist! but they also have a lot to do in 24h a day!)

IMO, nothing prevent you from studying after your work like if you are already a PhD student, give it a try: Take any advanced CS theory book or better, a Math book in your area of interest and work it out until you managed to do most of the advanced exercises. If you think this is not a healthy/sane life/work balance.. For a lot of people a PhD is not either. Take your time and be sure to understand everything in the book at its full extent. While you are doing this or after having done a couple of books, take high quality research papers in your field (lots of them are free), and try to understand them, implement them and try to reproduce their results. If you want a PhD in Deep Learning, do not expect that the PhD will consist in implementing a network, testing it, and start again. You may need to understand a lot of information theory, statistics, inference, math theory, etc. Also, for a lot of research level interesting ideas: implementing them is not possible in current frameworks, Expect to make a lot of coding. In any other research areas, usually (advanced) math is still a requirement: P/=NP kind of stuff in computability, graph theory, formal (type) logic (eg look at the rules for the Julia subtyping system[3] pg. 12), game theory (AI), category theory (for functional PhDs), stochastic calculus for finance, etc.. You don't need to know it all, only the part that are relevant to your interest, but still!

This may take month with long days to finish. but..

If you like this experience, you have two choices: 1) You feel that you are ready to do your research on your own, and in that case you don't need a PhD as you can continue to advance by yourself by taking any resources available. 2) You really need the diploma for whatever reason, give a shot at online universities, or try to find fast-tracks to obtain a master degree. Finally find an advisor that is willing to follow you and find funds for your PhD position.

In any case, after the PhD acceptance is where all the difficulties start: You need to study a lot, learn to master rapidly new material even if it is not the material you like, experiment a lot, report results, publish in high impact journals, review papers, read papers, prepare presentations, give presentations, give lectures, give exercise sessions, find fundings, write proposals, manage students, manage several urgent overlapping deadlines, prepare datasets, long hours, poor salary, etc. Don't expect that the PhD is related in any way to a bachelor/master program in my experience.

And don't think this will never apply to you if you find the right advisor, several studies study the link between a PhD and mental illnesses (~30%) [1][2], this is not exactly related to your capacity to work hard but more to the capacity of managing so many sources of anxiety while still being productive and often lonely.

All in all, this can be an extremely amazing experience or the worst experience in your life depending of your personality and ability to manage your time! I didn't write to discourage you to do it, but try in baby steps what it is like to be a PhD and after that decide for yourself. If you are convinced, find good advisors that are willing to follow you or guide you to a fast track that can suit you before applying!

[1] https://www.sciencedirect.com/science/article/abs/pii/S00487...

[2] https://www.nature.com/articles/d41586-019-03489-1

[3] http://janvitek.org/pubs/oopsla18a.pdf

somethingsome··on Terry Tao on some desirable properties of mathematical notation
I find it useful as he needed it for the following sentence to make sense :

A good notation should make this map Notation (and its inverse) as close to a (natural) isomorphism as possible.

Having said he want an isomorphism for Notation sums up all the following points he made explicitly (and more), so we could argue that we can keep the Notation and the isomorphism part but remove all the following point 1..9+

I think that in this way the post reach more people :) some needs only to read the first sentences, some the properties and finally some the example.

Isn't it a nice way to understand what he is trying to say in as much different ways as possible?

somethingsome··on Why Did San Francisco Schools Stop Teaching Algebra in Middle School?
Is it really a bad thing to not know algebra at the 8th grade?

I understand what you are saying, but I cannot reach the same conclusion as you that they are lowering the bar.

Even if at the moment a lot of people are exposed to algebra at a very young age, a very few people understand the maths behind, they use it as a set of rules to apply, which is very far away from how maths works..and then, leads to a bad use/generalization in unknown problem.

I didn't follow precisely the common core thing, but if I remember correctly, the idea behind is to teach to the child's to 'think' instead of having a rule based system. Ultimately leading to understanding algebra better and faster at a later stage, when it becomes only a corollary to one's thought.

When you can think more deeply and more abstractly at a young age, this may lead to better thoughts power, better logic, and ultimately better skills at math and in life. At least, I think this is the true idea behind the core, please if there is an expert on math education here correct me if I'm mistaken.

Of course, this is often difficult to accept for adults, as we want our child's to have 'at least' the same education of ourselves. But it is not obvious that the way we larned math, is a good way. If it is: why so many people do not like/understand maths? Isn't it a good thing to try newer teaching methods that we think are better?

I may be doing a big shortcut here: but from what I remember the core was worked out by very capable mathematicians in both mathematics and education and they were inspired by Korean/Japanese math systems that encourage the child's to think. If someone has references to this I would be interested as well :)

somethingsome··on H.266/Versatile Video Coding (VVC)
I don't think that neural networks are the right framework to achieve general purpose artificial intelligence (AGI). And indeed, the AI field may need a paradigm shift to achieve higher classification goals. I believe that probabilistic neural networks may be an interesting extension toward general purpose AI, even though this kind of networks need even more data.

If we take the example of a puppy, it seems to generalize pretty well using something like one-shot learning, but is it? I cannot confirm for sure how much data a puppy has already digested before being able to do what we could call "one shot learning". So maybe, the exposure to data is already there, waiting for a specialization toward a particular task.

Giving the ability to a network to be probabilistic enable it to do inference using uncertainty, which is clearly a neat feature when you are gravitating toward AGI for scene understanding.

In the case of video compression, scene understanding may introduce more artifacts IMO: Even if the scene is captured with high end cameras, on a pixel level basis, the edges will never be perfectly neat. I think this will decrease the ability of any network to "understand" which object is at the edges, this results in low classification rates on them, resulting in bad compression/decompression quality (?) for features that are important to the human eye.

All in all, I'm not sure that NN are the right tool for this kind of problems. But we are diverging from the main subject VVC, Thanks for the very interesting comments :)

somethingsome··on H.266/Versatile Video Coding (VVC)
I see your point, and I'm not particularly defending neural networks, but IMO, nothing prevents a network to generate a kernel able to detect 'fuzzy edges' and to refine it to an edge after some convolutions. So, if the input images are always consistent between them and with the input images for inference, I think the problem may be diminished (?), even if, as you say, we introduce some misclassification error. Obviously, to have the guarantee that all the input images are generated in the same way is a very strong condition difficult to achieve.
somethingsome··on H.266/Versatile Video Coding (VVC)
Don't you think that the iterated convolution process in neural networks is, in a measure, able to overlook this kind of 'visual trickery' ? I can imagine that the network is not able to perform well if you change the color profile of the input when you trained on another one, but small texture attenuations, diminished chroma components, etc. may not be as important when the image is downsampled and split a huge number of times (wondering)
somethingsome··on H.266/Versatile Video Coding (VVC)
It's surprisingly difficult to guarantee a high quality on every kind of videos and formats using a neural network. Furthermore, the network should be able to handle all the corner cases (think about color profiles alone..)
somethingsome··on H.266/Versatile Video Coding (VVC)
Usually several test sets are created periodically to handle any type of situation in a variety of resolutions and depending on which part of the codec people are working on, this process is very complicated and take a lot of time to ensure that all the kind of content is representative. The metric is not optimal for video content, but gives an idea if they are going in the right direction. Ultimately, blind tests with people (experts and non experts) are done on the content to have subjective measures added to the objective ones
somethingsome··on Sphere Eversion
I really like the mathematical impressions from Fomenko

http://chronologia.org/en/math_impressions/images.html

For programmable graphics, I like Penrose https://www.cs.cmu.edu/~kmcrane/Projects/Penrose/Penrose_SIG...

And finally the same author as Penrose (Keenan Crane) has a challenge for creating beautiful visualisation of abstract and difficult math concepts, but at the moment I'm unable to find the link

somethingsome··on Haskell for a New Decade [pdf]
Why would they be non-functional with today's technology stack? The whole stack is complex, but everything depends in general on very basic principles (eg: basic information knowledge about parallelism, how a compilation works, etc.. and hardware knowledge : how parallelism is handled, why do we need asynchronous jobs and how they work, etc..). Any complex piece of software can be decomposed into basic principles and very often without a lot of work. The only softwares that require more time to understand are the ones that solve some physical or mathematical problems IMO, as you need to understand the math underlying them. But again, this is not an impossible task
somethingsome··on Haskell for a New Decade [pdf]
It is not because you cannot name something that you don't understand something, even in profound levels.. the knowledge of the name is only conventional against the underlaying mechanism. It suffice to repair one wc, one time to have a pretty good idea on how it work without much effort
somethingsome··on Math-as-Code (2015)
You should take a look into Lean Theorem Prover (among others) ;)
somethingsome··on Show HN: Math to Code – Interactive NumPy tutorial for engineers
I was expecting some math but there is only a norm, I would be more interested in how to transfer from advanced math to code. example : what would be a good code abstraction for a Borel Space
← PreviousPage 9 of 9