New search-by-image method on Bing.com | Hacker News Reader

New search-by-image method on Bing.com

searchresearch1.blogspot.com

131 pointsrhema33 comments

These days I have same feeling about Google search results as I'd with Yahoo in late 90s. They routinely fail to identify important keywords and just ignore them. I have to put quotes on important keywords for roughly quater of my queries to avoid seeing generic stuff. For at least 10-20% of my queries, things are buried deep in to page 4 or 5 because rest of the pages are taken up by people gaming the SEO or they get ranked high because of stupid viral reason. Every person I have talked to about this has echoed this sentiment wholeheartedly. It appears that no one at Google cares anymore about all these pain people are experiencing. May be Google's internal metrics would even fail to capture this dissatisfaction because people keep using it despite of issues. Thanks to monopoly, traffic doesn't just drop. The image search by image is such an old ask that they have never got around to doing properly. Kudos to Bing for keep marching despite of our collective decision to just increase our pain threshold and stick with Google.

Good to see signs that Bing isn't just in maintenance mode, I'd hate to lose what little competition in search engines we have left. I switched back to Google from Yahoo after the latter got sold to Verizon, but it occurs to me that I never really did give Bing a fair shake. I suppose now's as good a time as ever.

I just finished spending two months on DuckDuckGo, and landed back on Google simply because I was tired of searching for so many things twice: once with the original search, and once with g!.

I really want a Google alternative. Maybe I'll try Bing again, too.

I've used bing for over a year now, rarely ever going back to google. I'm in the USA, no reference for outside the US...

Try Startpage. It proxies Google search results -- basically Google without the tracking.

And how do you know they don't track you? I have a problem with anyone that isn't willing to open source their code.

There is this thing called General Data Protection Regulations for European companies. You can just ask for the information stored regarding you. People have. That's how it is known they aren't tracking.

In other words, they have 30 days to erase your data. So, they can keep customer data for 29 days, sell it and then be in "compliance".

Startpage's published privacy policy goes much further than GDPR to state there is no logging of personal info.

They also seem to be voluntarily registered with the Dutch data protection agency that verifies this.

However none of this will matter to you if you don't a basic level of trust in what an online service states in their TOS/Privacy Policy etc

Bing is nowhere near Google yet, but I've been using it due to Google rate limiting my VPN. It does seem to pick up answers to programming questions quite well on the usual sites like SO, every now and then I'll have to switch because Bing seems to not index as many pages, and it seems to index at less frequent intervals to Google. Another thing is Google just seems to understand what you want even if you typed it poorly, Bing is bad at this so you end up typing more complete queries.

I've been using DDG for a few years now as my primary search engine. I would say 99% of the time I can find what I need (just like with Google), and it's super useful with some of their integrations into stack overflow and related sites.

Regarding the results, after a few years, I found that it's not that DDG is worse then Google -- but it's that I was conditioned on how to optimize Google's results. I believe it's because no search engine uses natural language, instead it's language that we think is natural but our brains have optimized for the engine.

Once I came to that realization, I started optimizing my search for DDG just as I had done for Google.

I often type a programming error message into DDG, fail to find anything useful, type it in Google, find it right away. I don't see how reoptimizing my brain would help in this situation where I'm just typing in an error word-for-word.

> I was tired of searching for so many things twice

Yeah I do this as a DDG user. I'm not even sure the google search is adding any value but I end up doing it "just in case".

I would like a way to get two columns of results for each search - one for DDG and one for Google, so that I can compare the two. It might wean me off doing the second search because I can finally resolve whether it's worth it or not.

there were some sites like this. I clearly remember googlegooglegooglegoogle.com that initially let you google in 4 frames, followed by a herd of clones where you could choose how many frames and which search engines to use where. and then there was something like googlevsbing.com that used 2 frames, left and right plus had synchronized scrolling afair. where are they all gone? this internet thing forgets its features faster than me it seems.

edit: also see https://www.makeuseof.com/tag/10-websites-to-compare-google-...

If you are in the US Bing works perfectly fine - I have used it pretty much exclusively the last year and don't miss anything. Some things I like better, like the changing fun backgrounds, clearer results, less ads...

If you are not in the US- it's not good... I have had bad results being abroad. Wish they would step up their game internationally.

Nice, though I use Bing daily (I am the only one, if I believe my collegues), I didn't know this exist.

I do wonder however, why the search function doesn't try searching on sub-parts of the image automatically.

Bing really has been improving compared to Google. I can find stuff on Bing which I can't easily find on Google because Google keeps rebuilding my search query, causing irrelevant results to show up.

Bing is my primary search engine. This feature has been in Bing for at least 6 months. I love it.

I use bing as my main search engine mainly for the rewards. I've gotten quite a few $5 amazon gift cards that way over the years.

However, it frequently doesn't return the results I need so I often go over to google.

Does it make you verify you're a human using a real phone number? I'm trying to go 100% sim free (Google Voice, Signal, WhatsApp) but Bing doesn't recognize VoIP numbers like Google Voice.

> I do wonder however, why the search function doesn't try searching on sub-parts of the image automatically.

i also wonder the same. But then it's quite a hard problem to automatically decide which part of the subimage to rerun a search (presumably, it's not free to run so many).

Actually this is a typical Machine Learning probleme, luckily was solved years ago and you can read about it here: https://medium.com/@ageitgey/machine-learning-is-fun-part-3-...

I don't think that is directly applicable to this case (definitely not to the level of calling the problem 'solved').

The link describes a technique for training a neural net to detect a type of image, an '8' for example, in an image regardless of location.

However, in this case, you only have one sample of the target. You could run a net to attempt to extract a list of features (children, bronze), and compare that to all your images, but I don't think that's how the bing image search is structured. That would be closer to searching for '4 children bronze park' (which it does a great job of finding such things; just nothing related to the target image).

Also, when you use convolution in the way described in the article, you would lose relative position information (or at least, dampen that part of the signal) of each of the bronze statues; as the entire point is to not care about the position of the object.

Geez, I didn’t realize Google and Bing had search by image. I’m still using tineye.

There are a good number of search engines for reverse image search, I've made a browser extension to search them all.

https://github.com/dessant/search-by-image

Very cool, especially the cropping feature! If anybody is curious as to how you would even start to tackle the problem, I wrote a tutorial to build your own prototype here: https://blog.insightdatascience.com/the-unreasonable-effecti...

This reverse image search engines are all really fascinating (Google, Tineye, Bing, Yandex). Recently I use more and more yandex.com reverse image search because it let's me find more associated pictures of an uploaded one. Side note: It's not censoring my searches... I'm an adult and I don't need supervision from a search engine company!

Not new. But yes useful when needed.

Yeah, I think this has been around for at least a year.

For others reading this, if you're looking for long tail content (e.g. running a ∃ type query), my experience with reverse image engines is roughly:

Yandex > Google = Bing > Tineye

For niche content also worth checking KarmaDecay and SauceNao if relevant. Bing's gotten a bit better recently, and with the big boys playing in the space Tineye's quality has definitely dropped off. As far as I can tell Baidu's reverse image search is just complete garbage and never has any useful results. I rank Yandex higher than Google not because their algorithm is better (although it does seem to be a bit more tineye-style forgiving in terms of editing), but also because they seem to greedily index images they crawl in a way that google doesn't (e.g. AFAICT a lower percentage of the images available in google's text-based image search index are available for reverse-image searching than on yandex)

For non long-tail reverse image searching (e.g. I just want a larger size or to find the original source) I mostly stick with Google though, because they seem to make it easiest to binary search the search timeframe like how they caught DPR.

If you're on Firefox (might be a chrome version avail now too?) Image Search Options is the extension you want so you can compare the different engines with a single click.

If others have good newer resources I'd be interested to hear about them.

Great comment, useful info! What is that about DPR? I had to do an acronym search. Dread Pirate Roberts? Is there a good story behind this about how they caught the suspect?

If I remember correctly Ulbritch was originally caught by doing basically a "git bisect" on the records associated with one of his usernames: https://www.nytimes.com/2015/12/27/business/dealbook/the-uns...

Thanks.

Does it require public URL like Google or is it based on classification?

it seems weird in this era of Machine Learning that the feature here is just an easier way for the user to experiment with different croppings of their image until they get good results

Reply on news.ycombinator.com