tomkarlo

I was thinking the same thing, but for sets of images where the occluding items are a relatively small percentage of the image area (and are moving around enough between frames), taking the median pixel value is effectively the same thing as the mode, but faster. (e.g. if you take 10 frames, the 1-2 frames where a pixel is occluded will almost certainly be an outlier to the 8-9 frames where the pixel is almost exactly the same, and the median will take the value in the middle of those 8-9 frames.

(Another problem with mode is that you'd have to posterize the image to ensure that shifting light, noise and camera movement don't cause the values of background pixels to vary slightly. Mode is much more brittle in that respect.)

If I recall correctly, Lightroom / Photoshop has a handy "median" filter, but no mode filter, which is why the popular method uses median.

It's like hollywood and their summer tentpole blockbusters. The need to make a huge movie that appeals worldwide discourages studios from straying from the most formulaic pattern available, because anything new is unlikely to have the same kind of broad appeal. Meanwhile, the huge budget (and high expected revenue) of those films makes the upside of making something better (because everyone's going to go see "Godzilla" or "Xmen" anyway.)

But people making lower budget films can take more risks, because if they do come up with a new type of hit they'll see huge returns.

I used to work on a site that covered the movie industries and one of our views was that fixed ticket prices were hurting the movie industry. A dumb, low-risk summer movie makes the same per viewer as an intelligent but niche film. Things would be a lot more interesting if some films targeted at smaller markets could charge $20 or more per ticket; and you might see declining returns on dumb tentpoles if they had to compete on price with each other (e.g. "How to die in the west" for $5 vs "Godzilla" for $15 ... ?)

It doesn't seem like that necessarily conflicts with native apps dominating smartphone usage - there's a significant number of use cases where a user wouldn't use an app, such as reading news articles linked from an app like Feedly. Mobile use is growing fast enough (and desktop is flat to shrinking) that a doubling of web usage may just represent the overall shift in usage patterns (i.e. desktop consumption of web moving to mobile.)

Slide 16 supports the apps over native story: app revenue is dominating browser advertising on an increasing basis.

SF isn't in Silicon Valley. San Jose is one of 10 largest cities in the county and Oakland's pretty big too. I'm guessing that "San Francisco metro area" probably can't include either of those two larger cities, nor would it include Silicon Valley (which is 15-20 miles from SF and closer to San Jose.)

It's not that different, depending how far north you get in Manhattan. The distribution of housing types might be different, but things get pretty residential in the northern end of Manhattan, and you're as far away from "midtown" than those places are from downtown SF.

In SF, some of the immediately adjacent areas (connected by subway) are drastically cheaper, like South SF, Oakland and Alameda. Like 50% of the cost if you look at the median sales prices. There are definitely some expensive suburbs, but they're either to the north in Marin or much further south down the peninsula down by Stanford.

It's worth noting that after New York and Brooklyn merged (I think around 1900), SF considered merging with Oakland.

Sure - it's like having a stopover at SFO and saying you've "been to San Francisco"... or at CDG and saying you "visited Paris." They're both 15 minutes from the city, too.

I kind of doubt anyone visits Oakland or San Jose (both of which are huge destinations in "the Bay Area" but have their own international airports) and says they "went to SF." I sure hope not. It's pretty clear cut if you went to SF or not, since there's water on 3 sides of the city and not much to visit on the 4th side.

Given that something like 60% of US households have two incomes, you'd have to expect that home pricing is at least somewhat tied to what you make on two incomes, not one. (And that's why you usually see these ratios in terms of median home price vs. median household income, but everyone immediately thinks of it in terms of per-capita income, which inflates it.)

This says more about the arbitrary nature of using city boundaries for studying home pricing than anything else. San Francisco certainly has high prices, but it also has one of the smallest footprints of any major city, and has lots of nearly-indistinguishable residential areas that are adjacent but fall outside its boundaries, particularly to the South and East. If you look just at pricing within the city boundary, it's going to be inflated by the fact that such a large percentage of that area is still relatively close to the "city center" and benefitting from the value that creates.

Meanwhile, "New York City" (which ranked lower on median price) includes large residential expanses that are much further away from the city center, including Staten Island. (Not to mention Los Angeles, which is what, 50X the size of SF proper?)

I'd like to see this comparison done for say, Manhattan island vs. San Francisco - that would be a much more apples-to-apples compare.

Everyone else besides the selected students.

You won't have created any kind of scalable / repeatable model that would work at the regional or national level; nobody is going to raise taxes enough to fund that and most people aren't going to sign up to send their kids off to boarding school for K-12.

That's a fairly common anti-fraud pattern for lots of systems: if a particular account's performance exceeds its long-term trend by a large amount (say 100% or more), it gets flagged for manual review. Once a payment goes out, it's not economic to retrieve it, so you want that extra review to happen before it's issued.

Ideally this would cause a delay in the payout, but if an account doesn't break the threshold until just before the payment date, the review might cause a delay.

This is common on ad platforms, credit card processors and even auction platforms like eBay. I would guess that it's a "standard practice" for risk mitigation.

The scraper bots are exactly why they're making the full information available to non-logged in users, so it shows up in Google's index. I believe it's considered a no-no to show the bot a different page from a non-logged-in human viewer, so they don't really have a choice.

Hiding it from users with login/session cookies, however, is just being annoying and depending on the fact that most people won't know enough to switch to an incognito window.

Yeah, that piece. It was great, but it also seemed like they're all doing business the same as in the 60's, at least from an operations standpoint.

I think this is closer to OrderAhead, which is an app that (at least in SF) a lot of restaurants use to allow you to place an order and pay for it, then just pick it up without having to wait on a register/checkout process.

(My one complaint about OrderAhead - they split any tip you give with the business, rather than just giving it all over to the business for the employees. That seems unfair to the employees.)

Car dealerships are apparently still pretty primitive in terms of inventory management - which I guess is unsurprising given that they're mostly small businesses. NPR recently did a piece on one of the top Jeep dealerships on Long Island and they had issues like selling the same car twice, or selling a car that was no longer on the lot, because they were literally using a whiteboard to track what had been sold (and even more informal methods for tracking cars that had been "put on hold".)

You want transaction volume, not total value outstanding, to help a market absorb larger transactions. The impact of a particular trade on a market is determined by the amount of trading going on (in particular, the size of the unfilled bids and the depth of the bids -- how much is available to trade, and how far away from the current market price is each potential trade.) At its most simple this is represented by the bid/ask spread, the gap between where you can buy and sell at a given moment - in more liquid markets, this gap is smaller.

Generally, a market where there are lots of transactions going on is going to have a bigger set of bids and asks right near the most recent transaction price. If a market is thin, than a large sale, for example, might have to chew through a lot of bids and push the price way down to get done.

Great and all (and I'm a big fan of smaller, lighter passenger cars) but it only works to a point - a tiny car might not kill a pedestrian or someone else in a tiny car, but it's unlikely to do well in an accident with a semi truck, or a large fixed barrier. (And people buy cars based on how well they protect the passengers, not others.) We should have smaller cars, but with reasonable safety features.

Fedex drivers aren't. That doesn't matter as much when there aren't real alternatives, but it can become critical if technology starts to drastically reduce the need for labor. That's part of what happened to the American auto industry during the 80's.

The fact that it started to roll through the terminal 2 drop-off (with me in the back) would indicate otherwise.

Yeah, but it's still added into the charge, right? I don't really worry about the components of my cab fare (including the tip), just the total.

The amount really isn't minimal when you consider it's a 8-12% surcharge on the typical UberX fare. Weren't they just doing a bunch of marketing about how they reduced their UberX fares 10% in SF?

The last time I took an Uber to SFO, the driver stepped out of the Prius without pressing the "Park" button. Maybe I would have gotten my $1 back.

Phones and small consumer electronics (even routers) have a relatively short operating life. It's a much bigger problem with systems that have a much longer expected service life and higher replacement costs, like cars, refrigerators, HVAC, thermostats, etc.

It's one thing to express and/or support a political view that others may disagree with, within the accepted bounds of our democratic process. (You can disagree with the viewpoint, but the act of making the donatio itself isn't illegal or immoral.)

It's another thing to subvert our system of government via deception, surveillance and/or intimidation, either by taking action outside the bounds of legal limits, or working to redefine those limits to suit your own purposes. There's a fair argument that these two situations are not comparable, regardless of how you feel about Eich's donation.

There's still an expectation that a skilled navigator can work out position and movement by hand, using charts, speed and compass headings ("dead reckoning") and have those results reasonably match/confirm the information reported by more modern systems. Using knots makes this significantly easier than it would be if speed and distance were reported using statute miles or another unit that's not easily converted into change in Lat/Long.

Depending on whether a fund is still in its reinvestment period, they may also not be able to put that money to work - it may automatically go back to the LPs, who aren't generally very happy to just get their money back when they're expecting 25% or higher IRRs.

Here's a question: when you return the computer, will they review its condition for damage / mileage, the way a car dealer would for a lease return? And if not, what's the incentive to take care of it?

You'll pay for a shorter lease term with a much higher monthly payment. Depreciation isn't linear; both a car and a computer will lose value much faster during the first 6 months than during the second 6 months (much less months 12-18.) So if you lease for say, 6 months rather than 24, the rate has to reflect that higher depreciation (plus the risk and cost associated with reselling the item at the end of the lease, spread over a much smaller number of periods.) I could easily see it costing double the rate Apple charges for a 24 month lease to rent for 6 months.

Example: $2000 machine that depreciates to $1500 after 6 months, 1200 after 12 months, $800 after 24 (keep in mind much of the value loss happens when you "drive it off the lot" and turn it from new into used). $100 overhead on accepting the return, cleaning it up and reselling it to extract that value. 6 mo lease: ($2000 - $1500 + $100) / 6 mo = $100/mo 12 mo lease: ($2000 - $1200 + $100) / 12 mo = $66/mo 24 mo lease: ($2000 - $800 + $100) / 24 mo = $50/mo

(Yes, I"m ignoring interest here. But rates are low.)

I've been using "Hello" and it's quite nice as a basic / super-simple SMS app. Very minimalist, though.