Hacker News Reader

The main application of background removal is online meetings and its already well implemented but it leaves a lot to be desired.

That's the most obvious case people see right now because we're doing more video calls in lockdown and people see it as a funny gimmick to play with on Google Meet, but background removal has been around for years in the professional video and film industry. Video chat is absolutely not the main application of the technology.

The video chat market though would be approximately a gazillion times larger than the professional video and film industry wouldn't it?

I have no idea. There's a gazillion times more people doing video calls than make professional videos, but not all of them would be willing to pay for the software and those who are probably wouldn't pay very much. It's the age old question of whether or not it's better to sell something cheaply at high volumes or more expensively at lower volumes.

Good webcams are hard to find at the moment because people are very willing to spend money on their video calls.

> people are very willing to spend money on their video calls.

MORE people are willing to spend money on webcams than in the recent past. That is why there is a supply issue. This doesn’t necessarily mean a huge number of people are willing to spend money in it in general. Doubling a relatively small number doesn’t automatically make it a big number.

>the age old question of whether or not it's better to sell something cheaply at high volumes or more expensively at lower volumes.

Factor in the support you need to accommodate your customers and you have your answer :P

You can just ignore support entirely and then it costs nothing. It's what Google does.

Better yet, create a "Help Center," that has 1,900 articles that all have a link that says, "Need more help?" and then that link goes to another of the 1,900 articles that also has a link that says, "Need more help?" And that link goes to another article, which has a popup that reads: Call 1-800-ABCDEFG. Then they call that number and get an automated system that says, "Thank you for calling. How can I help you today? You can say things like, 'I need help,' and 'I need different help.'" And then they have to speak all the menu choices. And the 0 button doesn't take them to a representative, only back to the first prompt.

This is too much fun to brainstorm.

Pretty hard to compete with the free implementations in most video chat tools...

What about marketing to live streamers and youtubers

Surveillance video processing and intelligence services are probably by far the largest use cases for video background removal and similar techniques. Far smaller in terms of number of customers, but far far larger in terms of revenue specifically for this capability.

Millions of people paying peanuts (attributable to this feature) is way less than a few hundred or thousand research / defense labs, contractors, etc., paying tens of thousands up to even millions on licensed software for this type of thing.

In both teleconferencing and surveillance you are dealing with static background, no need for fancy ML algorithms.

No, not true. Surveillance cameras mounted on police cars, helicopters, etc. Even in stationary cameras, scene understanding is a hard problem. Suppose someone wheels a dolly of boxes into a warehouse and leaves them. Are they now part of the background? Are they suspicious boxes? The coarse level state transition of objects in the scene is really hard to solve even with heavy ML.

My rule of thumb for advice on the Internet is that if it is mostly in the form of statements like "You need to do X, Y and Z" and has very few questions intended to dig deeper into the nuances of the product and business model... then it's not generally worth much at face value.

Essentially: The less information a person needs to dish out advice, the less value I am going to place on their advice.

haha you said what i wanted to say in a way more concise way. nice one

> Once you get traction, offer premium background images or videos. Create a marketplace where people can buy/sell.

A lot of this seems like guesswork, and honestly also like terrible advice. Maybe if the person had added their experience or a story about their own experience. Like the other person wrote, some advice might be irrelevant because this person hasn’t understood the product to know it’s it’s for existing footage only, not live footage.

Sometimes advice can be useful, but be wary when it sounds useful at first glance but includes no examples, and no credentials.

Then often it’s someone who considers themselves a success guru, without much sincerity. They just think that their advice is obvious, when its probably outside their domain. It’s not bad faith, it seems to me they’re just a wannabe startup guru/investor/founder/whisperer (what I write here might also be crap).

The best advice I’ve found comes with intimate stories of failure. Stories that help you understand the underlying friction, yet aren’t specific to the exact challenges you’re facing, but leave you with solid hypotheses to explore.

There is a lot of negativity in your response without even deluding to what's really wrong with the advice. There are tons of very successful founders recommending focusing on adoption instead of revenue in initial period. Freemium is very well adopted all over. Literally entire app industry worth billions of dollars use this model. Again, I'm not claiming to be expert in this domain. You are entitled to your opinion and frankly we are all armchair pundits here, however, it's bad form to spew out so much negativity without any reasoning or even alternative.

Yeah I understand where you’re coming from, and maybe my reply broke my own rule of giving advice without a personal story.

Where i’m coming from is having listened to advice from people who I looked up to, but who did it to get rid of me (because I was ignorant of protocols/etiquette), rather than being genuinely interested in my project. I’ve learned that the SV world has a high rate of failure and many are hungry, and they try to also carve out a role for themselves, or try to get involved somehow, even if that means they’ll claim competence in an area they’re not competent in. I was more naive then, compared to today, but believing these ‘gurus’ was a painful learning experience for me, because their advice often left me stranded in the end.

I’m not negative, it’s more just sharing my disappointments and struggles. Thanks again for your comment, which made me see I kinda broke my own rule

> Again, I'm not claiming to be expert in this domain.

Actually nowhere in your original post do you include this "not claiming to be an expert" disclaimer.

As a result, when I read your original post, I assumed that perhaps you were an expert as there was no self-doubt at all evident.

I think perhaps that's what the allegedly negative post that you are responding to is getting at.

>There are tons of very successful founders recommending focusing on adoption instead of revenue in initial period.

As of 2020, this is the exception rather than the rule. The problem is that focussing on adoption before focussing on value is very risky. You are constantly flirting with premature scaling before having found product/market fit (see e.g. Startup Genome project).

However, successful businesses that have been built on adoption are by-design much more likely to be well known. Again, it's entirely possible to build a business that relies on network effects or economies of scale, but it's not the default advice I would give to founder.

>Freemium is very well adopted all over. Literally entire app industry worth billions of dollars use this model.

No, you are confusing freemium with advertising here, which is the predominant business model. The challenge with freemium is the conversion into paid, which you cannot predict unless you have proven the value hypothesis already.

> I've personally tried half dozen solutions and all fell short in low lighting scenario.

Have you tried https://www.remove.bg ? I came across that site a while ago and was surprised how good the results were that I got. YMMV, though.

[EDIT]: Looks like Unscreen is actually from the same people as remove.bg. [0]

[0] https://news.ycombinator.com/item?id=24058701

We've built Unscreen on top of the tech of remove.bg, at least partly, and the #1 design goal for both products is a high result quality. We've summarized a few of the improvements recently here, if you're curious: https://www.remove.bg/b/introducing-remove-bg-x2

remove bg/unscreen are pretty much cool ML applications looking for a market

Reading “The Lean Startup” by Eric Ries.

He argues that this kind of “build it; they will come” aka “Just Do It” business process is a recipe for failure because you make a lot of untested assumptions about your market, and front-load a ton of (often wasteful) work.

He suggests the “Build, measure, learn” feedback loop, which you actually plan backwards:

1. Figure out what you need to learn by outlining the key assumptions, including your value hypothesis and growth hypothesis.

2. Decide how you will measure your success in validating (or invalidating!) those assumptions with science (eg he calls his method “innovation accounting”).

3. Build the smallest experiments (or MVPs) that will allow you to learn the things you need to learn.

4. Make adjustments to your hypothesis, rinse, and repeat.

Another common theme, “get out of the building” and actually speak with all kinds of potential customers. Since a start-up operates in an environment of high uncertainty they cannot rely on traditional planning.

Anyway I love reading the book and learning about the early days of Intuit, Dropbox, etc and how sometimes the way to make the most progress feels very counter-intuitive.

It almost feels like a real world version of Stochastic Gradient Descent for businesses to mechanically gravitate to various local minima on an invisible failure landscape.

The price and business model is perfectly fine:

1) most pros use green screens for background removal

2) for (rare) shots where background needs to be removed and green screen cannot be applied, a high quality removal like the one here is needed

3) for such rare occurrences, charging higher amount of money is completely appropriate and no reasonable pro would complain as it would save them additional costs of re-shooting the scene completely or doing the removal manually

4) nobody has anything similar, so they can charge premium while it's possible

Given what I mentioned I'd say their price is too low anyway.

Also recommend "The Mom Test".

There is a reason background removal sucks on chat apps. I fiddle around with the state of the art approaches, it takes a very expensive machine to get to an acceptable frame rate. It wouldn’t scale.

Would this be a good candidate for P2P? As what if your computer GPU could be used by other app's users to do a small part of such heavy operation? (when your GPU is not doing anything else)

Latency.

Can you elaborate more? What are the approaches?

Not OP, and not in this field, so do some more research, I am just sharing what I found:

https://paperswithcode.com/task/video-background-subtraction

I'm pretty happy with XSplit VCam (https://www.xsplit.com/vcam)

Anything like this for macOS?

According to the FAQ this only works with prerecorded video, not live video.

This looks like it's targeting pros like me who need to remove background from high-quality videos. The price is completely OK.

> Offer a desktop app that creates a virtual camera that you can use in Zoom, Teams etc. Differentiate yourself by adding features such as much more cool and fun background wall papers or even live videos.

https://www.xsplit.com/vcam does exactly that already and is available on demand or for a lifetime license of $40. It works mostly ok. Unscreen's quality seems a lot better though.

The very first thing I thought when I saw this wasn't "ooh, cool tech for voice chat", it was "man, the special effects industry will KILL for this - no more manually going through every frame of an effects shot and painstakingly removing background elements!"

So basically, I think the main assumption on which you built the rest of your advice is flawed.

The quality level here is way beyond what Zoom has or needs. This is a professional tool for creative pros — knocking out an image background is too easy but video is a real challenge. And this kind of work at this resolution would be prohibitively expensive to do for live video ... you’d need a Titan RTX or similar.

Sounds like that would evolve to become Google AdSense for your home office.

I don’t think it’s feasible to do this level of processing in real-time _for free_.

Pre-existing example: Snapchat has an app to apply filters in webcams

Right, this seems like a feature, not a product -- and it's not difficult to replicate. ML is becoming more and more of a commodity these days.