Jump to content
Sign in to follow this  
kcdeec

(Archived) Dumb question about image OCR

Recommended Posts

So am I incorrect in thinking that Evernote should be able recognize and index characters from photos i take using the iphone app? I mainly wanted to use it to catalog wine bottles and, say, could type in "syrah" in the search and it would find every image of a syrah I have, but it doesn't seem to work at all. Help?

Share this post


Link to post

You have to synch to the servers, wait for the OCR action to take place and then re-synch again. The time it takes to have the OCR data sent back to your client depends on the size of the image queue and whether you are a free or premium user (premium users are handled in preference to the free users. As there have been issues over the last couple of days for a lot of folk synching to the servers, there may well be more of a delay than usual right now.

Remember too, if the image is too fuzy then the note might be unintelligible even to the pretty amazing EN OCR software.

(actually it isn't OCR like you might understand OCR. The EN server software builds a tree of the possibilities for each word, and then that data is sent back to your client and associated with the note. This means that if it isn't sure whether a word was herring or hemming, both of those possibilites are associated with the note. That means you'd find the note by searching for either herring or hemming.)

Share this post


Link to post

is there a place to see where you sit in the processing queue for this? i downloaded evernote today to try it out, because i was kinda intrigued by the whole image text scan feature. i've spent the last 6 hours trying to figure out why none of the images i was loading in, either by desktop clippings or by iphone, were not picking up the text.

wholeheartedly frustrating to find out that there's some stupid queue mechanism, which clearly has extended delays, which clearly makes this application near useless to me, particularly when there's no discernable way to see that your images are indeed being processed... eventually.

Share this post


Link to post

Queues for the free accounts are backed up according to the status page.

http://status.evernote.com/

The queue backups seem to be a fairly frequent event right now so I too would like to see a dashboard indicating current (updated every 1/2 hour or so) email and OCR queue times on the status page.

Share this post


Link to post
is there a place to see where you sit in the processing queue for this? i downloaded evernote today to try it out, because i was kinda intrigued by the whole image text scan feature. i've spent the last 6 hours trying to figure out why none of the images i was loading in, either by desktop clippings or by iphone, were not picking up the text.

There's no place to see where you are in the queue, and if you're a freemium user, it wouldn't make much sense anyhow, as premium users get first priority.

wholeheartedly frustrating to find out that there's some stupid queue mechanism, which clearly has extended delays, which clearly makes this application near useless to me, particularly when there's no discernable way to see that your images are indeed being processed... eventually.

I'm curious to know what mechanism you'd use that's not a queue.

~Jeff

Share this post


Link to post
I'm curious to know what mechanism you'd use that's not a queue.

~Jeff

well lets see... you have thousands of beefy PC's and Macs that are likely running the desktop client, right? why not run a check to see if the desktop client is active for a user and offload the OCR function (which really is a key feature, no?). it could be preprocessed at the client-end before uploaded to the cloud, as the cloud obviously can't keep up.

the snipe about being a freemium user is total *****, too. i'm a freemium user because i want to see how this thing works. aaaaand it doesn't. why the hell would i go about spending $5 a month on it? from what i can tell, premium users are having a bunch of problems too.

Share this post


Link to post
well lets see... you have thousands of beefy PC's and Macs that are likely running the desktop client, right? why not run a check to see if the desktop client is active for a user and offload the OCR function (which really is a key feature, no?). it could be preprocessed at the client-end before uploaded to the cloud, as the cloud obviously can't keep up.

You've evidently not followed the recent problems that they had with their servers, which have been addressed and are just now clearing out. Delays of that magnitude are rare here, in my experience; they're typically much smaller than what folks have experienced recently (there's plenty of threads that talk about it, if you care to search for them). The OCR feature is something that they prefer to do in the cloud; not every client machine is 'beefy' (I pretty much do NOT want my netbook doing OCR, thank you very much), and having OCR in the cloud ensures that all platforms get the same OCR experience. There may be other reasons that Evernote staff could chime in on, but given that the system is designed for server-based OCR, a queue is the fairest way to go

the snipe about being a freemium user is total *****, too. i'm a freemium user because i want to see how this thing works. aaaaand it doesn't. why the hell would i go about spending $5 a month on it? from what i can tell, premium users are having a bunch of problems too.

That wasn't a snipe, it's a fact -- it's Evernote policy. There's nothing wrong with being a fremium user, Everntoe encourages it, and it's very useful as-is. Going to premium gets you some more features, more storage and better support, and priority in OCR. Not sure what the problem is; if you don't want to go premium, that's fine too.

~Jeff

Share this post


Link to post
the snipe about being a freemium user is total *****, too. i'm a freemium user because i want to see how this thing works. aaaaand it doesn't. why the hell would i go about spending $5 a month on it? from what i can tell, premium users are having a bunch of problems too.

Not sure why you think that's total *****. As Jeff said, it wouldn't really serve much of a purpose for free users, since premium users get priority. A free user may get to be 5th in line when one or more premium users add a bunch of notes that then bump the free user down to 85. Then you get back to position 25 & some more premium user come along & add more notes & bump you back down again.

Share this post


Link to post
You've evidently not followed the recent problems that they had with their servers, which have been addressed and are just now clearing out. Delays of that magnitude are rare here, in my experience; they're typically much smaller than what folks have experienced recently (there's plenty of threads that talk about it, if you care to search for them).

yah i'm familiar with them. the whole 85% capacity nonsense, and delayed proactivity along with poor planning for unexpected demand a la mac app store launch. doesn't give me a warm n fuzzy on the 'mission critical' front. if i ran my shop like this... mass exodus.

take it from a brand new user's perspective... and there's quite a few of us as per EN's own public 'hoorah' about how many newbies came on board recently. this initial experience does not bode well for continued use of the app. particularly when there's zero in-client status information - i had to go hunting for it. imagine how many others are finding this to be quite irritating from a first-use perspective.

as it stands, i'm still waiting for a couple test images to provide me with searchable text.

don't get me wrong, i'm not all negative nancy. i just expected a lot better when all i ever hear about is how awesome EN is. if i can prove the use case and demonstrate that it works as advertised, i'm likely to purchase a lot of licenses for my company. that said, i'd have to see a lot more servers thrown at this thing to ensure that it's a lot more reliable. otherwise dropbox is still my proven sync tool of choice, plain-text or not.

in any case, appreciate the commentary to date. i'll pipe down and watch from the sidelines now.

Share this post


Link to post
Not sure why you think that's total *****. As Jeff said, it wouldn't really serve much of a purpose for free users, since premium users get priority. A free user may get to be 5th in line when one or more premium users add a bunch of notes that then bump the free user down to 85. Then you get back to position 25 & some more premium user come along & add more notes & bump you back down again.

yep, that's totally fine. but _some_ form of feedback to the user is critical here. otherwise they are lost in the app, and are even more lost as a continued user of EN.

i'm savvy. i get it. i can figure stuff out and understand it. but jeez... there's no way my dad would figure this out. or my university-girlfriend for that matter. they'd immediately complain that it doesn't work and then stop using it altogether because of no direct user feedback from the system that says "hey, we got your note/screenshot/picture/pdf... give us approx "x-time" to get this thing scanned in and sync'd back to you"...

Share this post


Link to post

well lets see... you have thousands of beefy PC's and Macs that are likely running the desktop client, right? why not run a check to see if the desktop client is active for a user and offload the OCR function (which really is a key feature, no?). it could be preprocessed at the client-end before uploaded to the cloud, as the cloud obviously can't keep up.

There's more than one way to skin a cat. I'm a premium user and I am doing exactly what you suggested. I have my handy-dandy miniature Fujitsu ScanSnap scanner set up to automatically OCR everything. It adds about 30 seconds to each scan. I never have to think about OCR. As soon as it is imported into Evernote, I have full access to all of the programs features.

Share this post


Link to post
in any case, appreciate the commentary to date. i'll pipe down and watch from the sidelines now.

I have to agree, anyone who hopped on in the midst of their server woes was not going to get a good first impression. That's unfortunate because there's a lot to like about Evernote, and even the free version is pretty useful. Doesn't mean that there's not room for them to improve, though. Anyhow, I hope that your images come through shortly, and things go better for you from here on out. Good luck.

~Jeff

Share this post


Link to post
yah i'm familiar with them. the whole 85% capacity nonsense, and delayed proactivity along with poor planning for unexpected demand a la mac app store launch. doesn't give me a warm n fuzzy on the 'mission critical' front. if i ran my shop like this... mass exodus.

I am right there with you on this. Evernote has graciously refunded my money and I'm on the free train once again.

Share this post


Link to post

well lets see... you have thousands of beefy PC's and Macs that are likely running the desktop client, right? why not run a check to see if the desktop client is active for a user and offload the OCR function (which really is a key feature, no?). it could be preprocessed at the client-end before uploaded to the cloud, as the cloud obviously can't keep up.

For free users of Evernote, it might be actually a good idea to have the processing done on their own client if it's feasible; For example, a minimum system spec would be established, OCRing at idle times, ...

For Premium members, the workload would be then offloaded to the cloud for OCR indexing. That would be an interesting solution to combat the current experienced problems.

i'm savvy. i get it. i can figure stuff out and understand it. but jeez... there's no way my dad would figure this out. or my university-girlfriend for that matter. they'd immediately complain that it doesn't work and then stop using it altogether because of no direct user feedback from the system that says "hey, we got your note/screenshot/picture/pdf... give us approx "x-time" to get this thing scanned in and sync'd back to you"...

Agreed. For more patient users who have to wait for the queue, figuring things out would be a solution. For those who aren't computer-savvy, they'll migrate to another service or not use one at all.

well lets see... you have thousands of beefy PC's and Macs that are likely running the desktop client, right? why not run a check to see if the desktop client is active for a user and offload the OCR function (which really is a key feature, no?). it could be preprocessed at the client-end before uploaded to the cloud, as the cloud obviously can't keep up.

There's more than one way to skin a cat. I'm a premium user and I am doing exactly what you suggested. I have my handy-dandy miniature Fujitsu ScanSnap scanner set up to automatically OCR everything. It adds about 30 seconds to each scan. I never have to think about OCR. As soon as it is imported into Evernote, I have full access to all of the programs features.

Alright... Let's print out each separate image we want to index, and then add it to the stack of image son the Fujitsu ScanSnap... Oh wait, I'd have to invest $250-450 for a scanner... And what if I want to index something that requires a flatbed scanner! Oh no! Fujitsu ScanSnap isn't one!

Share this post


Link to post

Alright... Let's print out each separate image we want to index, and then add it to the stack of image son the Fujitsu ScanSnap... Oh wait, I'd have to invest $250-450 for a scanner... And what if I want to index something that requires a flatbed scanner! Oh no! Fujitsu ScanSnap isn't one!

In my situation, I've had a flatbed scanner for years. Never use it, just collects dust.

Amazon sell price for my ScanSnap is $245 with free shipping

My two most active scanners are the ScanSnap (for just about anything) and Neat Receipts (for thermal receipts)

Share this post


Link to post

Any idea what the queue is looking like now? This is my second attempt at trying to use Evernote. The first time, I tried it out to see how well image indexing worked, but after a few days, I didn't see any results and quit. I figured I'd give it a second try, but it looks like I have bad timing and I'm stuck in another queue. It's been 3-4 days and I'm a free user. Any ideas on how long it might take? I took a sample picture of a wine bottle, a book, and a letter.

Thanks!

Share this post


Link to post

Thanks for the reports.

Processing of images and PDFs for Premium accounts currently starts within a few minutes. The actual processing of a PDF may take up to an hour, but we'll start working on it within a few minutes. Images rarely take more than 5 minutes from start to finish since they're only a single "page".

Processing of images for Free users is currently taking several days. If you take a picture of a business card and need to recall it next week when it's way down in your list of recent notes, you should be able to search for it. But if you just want to "test" the recognition as a free user by loading an image and then immediately searching against it to play with the feature, that's not working.

Unfortunately, the sudden growth in both users and images exceeded our capacity by more than a small amount. I.e. We didn't go from "80% capacity" to "88%" capacity in one week, but rather jumped from "80% capacity" to "115% capacity" in a single week. We had additional servers in IT pipeline for more gradual growth, but not this kind of massive spike.

We're working to make the processing of images for Free users more responsive so that you can play with it and demo it to see it working. We've increased the image processing capacity by about 25% in the last week, and are adding another five servers by Monday. This requires ordering high-end servers with 16GB of RAM and 12 processing cores, along with licensing for the commercial OCR package that is used for some of the work on those boxes. We're also doing watching the existing boxes closely to prod them to maximum performance while we're behind.

All of your images will get processed, and we will catch up on the backlog. Anyone who pays $5 to upgrade to Premium this month will jump to the front of the queue for their new images, otherwise we appreciate the patience of the non-paying users and hope to get the service level back where it belongs soon.

Thanks

Share this post


Link to post

Thanks, Dave. I'll wait another week or so and check back. In the meantime, is there a target wait time that Evernote strives for the Free users? It's clear that the target for paid users is a few minutes, but I couldn't find anything regarding the desired wait time for image recognition/indexing for Free users.

Share this post


Link to post
Thanks, Dave. I'll wait another week or so and check back. In the meantime, is there a target wait time that Evernote strives for the Free users? It's clear that the target for paid users is a few minutes, but I couldn't find anything regarding the desired wait time for image recognition/indexing for Free users.

On Nov 23, Dave said: "If a free user uploads an image to their account, it should be processed within 2 hours or so."

Share this post


Link to post

Alright... Let's print out each separate image we want to index, and then add it to the stack of image son the Fujitsu ScanSnap... Oh wait, I'd have to invest $250-450 for a scanner... And what if I want to index something that requires a flatbed scanner! Oh no! Fujitsu ScanSnap isn't one!

In my situation, I've had a flatbed scanner for years. Never use it, just collects dust.

Amazon sell price for my ScanSnap is $245 with free shipping

My two most active scanners are the ScanSnap (for just about anything) and Neat Receipts (for thermal receipts)

Ahh! But that's you. There are certain people who use and need flatbed scanners (Graphic designers, architects, students, cooks, librarians, ...) and not just businessmen (Or what occupation you are have.) scanning in receipts. Can't necessarily scan in a sketched design through a scansnap that might crumple the design or blemish the ink. (Again, the quality of using one of the 'roller' scanners varies.) Or scan in a few pages of a book that you intend to annotate.

Share this post


Link to post

In my situation, I've had a flatbed scanner for years. Never use it, just collects dust.

Amazon sell price for my ScanSnap is $245 with free shipping

My two most active scanners are the ScanSnap (for just about anything) and Neat Receipts (for thermal receipts)

Ahh! But that's you. There are certain people who use and need flatbed scanners (Graphic designers, architects, students, cooks, librarians, ...) and not just businessmen (Or what occupation you are have.) scanning in receipts. Can't necessarily scan in a sketched design through a scansnap that might crumple the design or blemish the ink. (Again, the quality of using one of the 'roller' scanners varies.) Or scan in a few pages of a book that you intend to annotate.

Yup, it is me and a lot of other folks. Sure, there are certain people who need the large flatbed scanners that take up a third of the deskspace. I have to question how many students living in college dormitories own full size flatbed scanners. And with the huge amount of free recipes, cooking software and apps, I don't see a growing demand for cooks. You do have a point with graphic designers and architects. Lawyers as well. There will always be a market for niche requirements. But the majority of consumers live in a 8.5 x 11 (or A4) world. The ScanSnap does a super job on resolution on those sizes.

After reviewing the products at the 2011 Int'l CES, which just wrapped up in Las Vegas, you have to agree the market direction is to smaller and more portable devices. iPad, tablets, Livescribe smartpens, e-book readers, mobile phones, etc. And I am finding more and more of the big paper stuff is available for download in digital format for the masses. Especially books, even text books. As I said in the beginning, there are some people who need the classic flat bed scanners. There are some people who still use faxes. I'm just saying... to each his own.

Share this post


Link to post

Our IT folks worked into the night to repurpose a bunch of hardware to do image processing since the set of OCR library license keys arrived before the next batch of boxes. This reduced the backlog by a few hours, but we're still several days behind, so will keep churning on them through the weekend.

Share this post


Link to post
Our IT folks worked into the night to repurpose a bunch of hardware to do image processing since the set of OCR library license keys arrived before the next batch of boxes. This reduced the backlog by a few hours, but we're still several days behind, so will keep churning on them through the weekend.

Oh gosh, I am just imagining a closet containing an old Amiga, a Next cube and a few TRS-80s, linked via baling wire...

~Jeff

Share this post


Link to post
Our IT folks worked into the night to repurpose a bunch of hardware to do image processing since the set of OCR library license keys arrived before the next batch of boxes. This reduced the backlog by a few hours, but we're still several days behind, so will keep churning on them through the weekend.

I'm not sure if I recall this correctly, but I thought Evernote developed its own OCR engine... Or is it only the handwriting recognition engine?

Share this post


Link to post

We actually use a combination of separate engines, and then merge the results of each engine together. Three of these engines were developed by Evernote employees at one point or another, and a fourth one is commercially licensed.

Combining results from multiple engines allows us to incrementally improve the overall recognition level for a word, since different engines have different strengths. The separate engines also give us support for more languages (especially Japanese and Chinese).

Share this post


Link to post

Just another update: in the last three days, we have been able to increase our image processing capacity from 86 CPU cores to 136 CPU cores to give us more than a 50% boost in our image throughput. We are now handling well in excess of one million images per day.

As of today, we are making significant progress against the backlog of images from Free accounts. The backlog is still significant (the oldest image in the queue is approximately 3 days old), but we are now reducing that at a steady pace. I hope that the combination of the new capacity and the lower load from the weekend should allow us to catch up by Monday.

Share this post


Link to post
We actually use a combination of separate engines, and then merge the results of each engine together. Three of these engines were developed by Evernote employees at one point or another, and a fourth one is commercially licensed.

Combining results from multiple engines allows us to incrementally improve the overall recognition level for a word, since different engines have different strengths. The separate engines also give us support for more languages (especially Japanese and Chinese).

Ahhh... Alright. That explains why the OCR system seems very demanding for the Evernote servers.

... and the lower load from the weekend should allow us to catch up by Monday.

... Shouldn't there be more load because its the weekend? More people using Evernote?

Share this post


Link to post
... Shouldn't there be more load because its the weekend? More people using Evernote?

No, our usage goes down a bit on the weekends. We get more usage on weekdays, primarily since more people are sitting in front of a computer at work rather than enjoying the great outdoors.

It varies by client ... Windows usage in particular takes a big dip on the weekend, but mobile doesn't as much.

Share this post


Link to post
... Shouldn't there be more load because its the weekend? More people using Evernote?

No, our usage goes down a bit on the weekends. We get more usage on weekdays, primarily since more people are sitting in front of a computer at work rather than enjoying the great outdoors.

It varies by client ... Windows usage in particular takes a big dip on the weekend, but mobile doesn't as much.

That explains it! :)

Regarding Evernote's storage architecture - How redundant is the system? Are the storage parts of Evernote outsourced to a cloud hosting provider (AWS, a CDN, ...) or stored locally? If a hard drive failed, which could happen any time, how much of Evernote would be affected?

Share this post


Link to post

all 23 of my pics were indexed by this afternoon. i was impressed by how well it worked and will definitely be keeping the app. hopefully the wait time for free users gets back down to just a few hours; i think it took about four days for me.

Share this post


Link to post
Guest
This topic is now closed to further replies.
Sign in to follow this  

×
×
  • Create New...