Cheezam – Shazam for Cheese | Hacker News Reader

Cheezam – Shazam for Cheese

cheezam.fr

254 pointssvenfaw104 comments

I got 15 cheeses from FR a few days ago (all French cheeses from the vicinity of Blois) and it had 100% of them wrong. Cheese 16 was a cheddar and it categorised that as Comte, but now I see here that it only recognises French ones? I do like the idea though; my wife can recognise many on sight and smell but I am more of a barbarian and know practically nothing about these except that I like or not like; it would be a handy app for me anyway if it performed better.

Funny, I just tried it on a comte and it thought it was gruyere. Still probably a better guess than a lot of people I know would give

(Edit, reading below I see others noticed the same thing)

Compté basically is Gruyère: same recipe, same process, same taste profile, just slightly different terroir.

(https://www.thephcheese.com/gruyere-and-comte-two-of-a-kind)

Fair enough, though the reason I posted was because the GP did get cheddar misclassified as comte so it evidently can propose both comte and gruyere as labels. Maybe the taxonomy could be revised to make it respond gruyere/compte if they are really indistinguishable.

Where I am, comte is usually wider than gruyere. If this generally holds, that could be what it uses to distinguish them, and actually I showed it a cut piece so maybe that's why it thought gruyere.

With more time / motivation, it would be interesting to do a study looking at whether size and orientation can change the predicted label. Like if I cut cheddar into a wedge would it think its brie?

Comté. Note that it is very interesting that to make good Swiss Gruyère, you need Montbéliarde cow milk. At the turn of the last century, farmers from France effectively passed calves across the border by carrying them on the back and doing very dangerous mountain climbing[0].

[0]: https://www.sac-cas.ch/fr/cabanes-et-courses/portail-des-cou...

It's a gold idea from an advertising agency. It's sadly just fluff for award shows.

No you cannot say that my Comté affiné 36 mois is a slice of gruyere instead.

My guess is it's a labeling issue -- Comté is sometimes confusingly referred to as Gruyère de Comté, though it is obviously not a Swiss cheese.

To me it seems much more likely that it is just unable to distinguish these cheeses. Both cheeses look very similar both in the cheese itself and the rind. They are also packaged and served in similar ways.

To be fair, Comté and Gruyere are pretty much as similar as 2 different types of cheeses can get.

visually, maybe. in taste, not at all. Comte and Beaufort maybe, Comte and Gruyere? no way

> Comte and Beaufort maybe

This is heresy!

It's hard to taste a picture! I'm not saying they're exactly the same, but they're made in extremely similar ways.

I'm from a less cheese-y part of Europe, but isn't Comté often called Gruyère de Comté for a reason? Of course they'll taste very different when aged for very different amount of times, but I've found them to be decently similar when aged for similar amounts of time.

I’ve never seen Gruyère develop those strong flavours (18m+), then again to be fair in France Gruyère is rarely aged

Oh. No. Really not. Comté is way more flavorful and tangy, and develop quartz of salt than crisp when you eat it.

The crystals are not salt but amino acid, mainly tyrosine.

Also calcium lactate, which is a salt.

Ok. They are pretty salty and transparent.

I've never been so angry because of an HN comment, nom d'une pipe en bois. But cheers from Switzerland anyway, let's enjoy cheese while we still have a sense of taste.

Non mais vous avez mieux que le gruyère pour rentrer en compète avec le comté non?

The crystals are just from aging it for longer, aren't they? Of course a 36 month Comté vs. 4 month Gruye won't have that many similarities compared to a 12 month Comté vs. 12 month Gruyere.

Correct.Yeah they are.

You specify the range of age you want when you buy comté.

For Gruyère, I never saw that, but it might be because I grew up a few hours from where it’s made.

To clarify : real gruyère is a great cheese. But comte is one of my favorite.

Man I miss cheese.

No, really really no. Please try again ( and make sure you get proper cheese, not tasteless plastic which will lead you to believe they're the same)

People are saying no, but nobody is offering a more-similar pair of cheeses.

beauford is much more similar to compté

Gruyère and Emmental.

Not at all. Emmental has a mild, flat flavour profile where Gruyère has a full bodied rich taste, just like Comté does.

That's weird, it simply says "petril" for every image.

It did say no cheese on some images I tested... And it was right.

Exactly what I first thought of. For context: https://www.youtube.com/watch?v=B_m17HK97M8.

“adding a new class, catastrophic forgetting, and retraining” is … exactly true for DNNs.

Nicely surprised, a picture of a Mont d'or was labelled as "Vacherin des bauges" which is quite close. Mont d'or is another name for a "Vacherin" from "Haut-Doubs".

Unfortunately, the /label page did not work on Firefox Android for me.

Reminded me of the Monty Python Cheese Shop sketch.

https://en.wikipedia.org/wiki/Cheese_Shop_sketch

“The original codename for Python Package Index (the central software repository for the Python programming language) was "CheeseShop" in reference to this sketch. The current standard packaging format for Python software is called "wheel", as a reference to a wheel of cheese.”

Of course, it's a `.fr` :D

Side note:

LOC for the example API call in

  JS:        13
  Python:     9
  Curl/Bash:  1
  Swift:     60

Where is this? 9 lines for Python seems excessive. Should surely only be two, one of which is an import?

In the "How does it work" section at the bottom. You're right, Python could be much shorter, it defines a few variables for prettiness.

It's very misleading that the author decided to use external libraries on the other examples, yet used the built-in URLSession on Swift.

The JS fetch example is native, no library. But point taken - what does a Swift file upload request look like with a (common) library?

You can find an example using Alamofire, the most commonly used network framework, here.

https://github.com/Alamofire/Alamofire/blob/master/Documenta...

My swift is pretty rusty now, but I'm fairly sure you don't need to construct an http request by hand (which is what the example code is doing), even if you're limited to only using Foundation classes

The issue is the multipart/form-data body. You still need to build it by hand in current Swift.

https://www.kaggle.com/mathurinache/cheese-pics

https://query.wikidata.org/#%23Subclass%20of%20Cheese%2C%20w...

Don't have french cheese around me right now, I tried with the first google image of camembert and it mislabeled as brie...

On a tangent, as I'm developing a similar ai powered pwa, what's, in your opinion, the preferred way to let the user shoot a photo and upload it? Leverage the native camera/upload UI (like the cheese app) or using a custom html/js UI without leaving the app (something like this [0] demo not mine)?

[0] https://demo.kasperkamperman.com/mobilecamtemplate/

edit added link to js UI example

To be fair camembert and brie look identical from the outside ( and even mostly inside).

In other words, there cant be a Shazam for cheese. It’s an impossible task.

My rule of thumb is: if expert human eyes can't classify a thing looking at it, the ai will also fail.

That’s not entirely true. For example, ML is used in radiology to find problems that are too subtle for the human eye to detect.

But pigeons still do it better: https://twitter.com/emollick/status/1388594078837878788

Maybe someone can invent a cheese tasting accessory for phones. Not sure if I want to share my cheese with my phone though.

Maybe if it was equipped with a spectrometer and microscope...

In this case it should be uncertain and return more than one result.

When I'm identifying a cheese I feel like it's 50% sight, 40% smell, 10% feel. It's hard to imagine identifying arbitrary cheeses with just sight. There are so many and a lot of them look extremely similar.

Even distinguishing between Jarlsberg and Swiss seems like it could be a challenge, and that would be easy relative to a lot of others. On top of that if you have something like brie there's a million of them - I don't think I could distinguish brie de nangis from another brie unless I could touch it.

I read "Shazam for Chess" which would be quite interesting.

1.) Make a photo of a chess position

2.) Recognize position and convert to digital format

3.) Search chess database of grandmasters and return relevant games in png format

Got the Saint-Nectaire! Bravo!

One of the best cheeses !

They called my cheddar a Pelardon. The barbarians.

News flash from 2023: Prevision.io rebrands itself as Cheezam to focus on the fine cheese and wine niche. The company has partnered with Michelin 3-star restaurants, applying machine learning to high-throughput sequencing data to offer menus specially tailored to your genetic taste signature.

How close are we to having a webcam/microphone for smells? Even if it only detects a small fraction of the thousands of the existing molecules, it would definitely be interesting, especially with some learning model built on top.

I found Sean Carroll’s Mindscape podcast episode on smell to be insightful on this[0]. My understanding? Not anywhere close.

0 https://www.preposterousuniverse.com/podcast/2020/07/13/105-...

Nice.

I have a generic blue mold cheese here and it identified it as Bleudauvergne.

Incorrect, but very close to the truth.

There are so many veined cheese, It's hard to classify them :/ Sometimes the model succeed thx to the packaging but maybe I should predict generic "blue veined cheese" instead :/

Hi cheezam developper here. I got som server alert this night, it seems it comes from here :D

If you got any question, just do

First thing first : I labelled the cheese by myself , self taught. So most of the errors come from me labelling

other points : as a french, I have access to french cheese for my photo for the most part, this i why the model is focused on french cheese

But I hope to learn more about other countries cheese ! We should probably put some image gathering / labelling website somewhere

As a child, I loved when we visited our family there: fruit as an appetizer and cheese for dessert was much more natural to me... "put a piece of cheese in the ground and you'll find out"

Silicon Valley on HBO was a prophecy.

This one's an exact Web3 sh*t prophecy: https://www.youtube.com/watch?v=nJVW_DqLntg

I’d really like a Pandora for cheese recommendation service based on current cheeses

I gave photo of Brie cheese and got vieuxboulogne. Well, they need to improve.

Is it it similar to Hotdog/Not hot dog app shown on Silicon Valley?

Reminds me more of cheesoid

https://www.youtube.com/watch?v=6Ofesgkhikg

I put in a pic of chedder and it says "gruyere". Nah bro.

That's because it detects only French cheese. It's not trained on British/Swiss/Italian/Dutch cheeses.

This is one of the least surprising TLDs I've seen.

It thought the kids Laughing Cow triangles were Etorki…

Reblochon is seen as Cantal (quite sure).

How huge is the dataset?

Zut alors!

Nailed it https://imgur.com/a/WQMr5w9

Detected all my cheddars as Gruyère…

What odor transducer do you use?

SeeFood

They only seem to be interested in French cheese.

What a pity... There are so many great non-French cheeses...

There are, but unless I'm mistaken no country has even close to the same amount of "great cheeses", so might as well start there.

Switzerland disagrees.

Canada concurs

Switzerland can disagree as much it wants, it doesn't make the fact less true.

The UK disagrees

UK produces more varieties. Of course it depends what "great cheese" is defined as - naming is a lot looser in the UK - but a cheese DB without UK (and definitely Italy too) is missing a lot.

Sure, but quantity does not equal to quality.

I doubt there are many of those named varieties that anyone would truthfully say are low-quality produce by definition, though. There is just a rich, long history of cheesemaking and a wide variety of different cheeses.

Developer here. As a french I only had access to french cheesemaker, so my database if majority of french cheese

You have to knwo that most French cheese make WONT sold cheese from other country. If you want italian cheese, you need to go to Italian cheese maker and if you wan cheeddar, you go to mark and spencer :D

Thus most of my dataset are very local french cheese

What about Venezuelan Beaver Cheese?

"great cheeses" is rather subjective... Wikipedia (and my personally biased experience confirms) states who has got the most cheese variety:

https://en.wikipedia.org/wiki/List_of_Italian_cheeses

I would argue that it is more probable to find more "great cheeses" where you have many different varieties.

France does machine learning

"Not a hot dog"

For those not getting the reference: https://youtu.be/mrk95jFVKqY

Not Hotdog (SeeFood Technologies Inc.)

What would you say if I told you there is a app on the market that tell you if you have a hotdog or not a hotdog. It is very good and I do not want to work on it any more. You can hire someone else.

https://apps.apple.com/us/app/not-hotdog/id1212457521

(Seller: Home Box Office Inc.)

Every time I think I've reached the end of the Internet again (first time, 1993) I am proved wrong.

That attention to detail... The IDE is "HooliCode".

"platform=hoolistore".

Very impressive.

They even set up legit GitHub profiles with repos that appeared in some scenes (and even their site/blog used to have regular updates lol). It was awesome.

My first thought.

Goddamn it Jian Yang!!!

Not a hot dog

Reply on news.ycombinator.com