Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    markkastoun
    @markkastoun
    okay great ... should I use pastebin for the logs or you have a preferred method?
    markkastoun
    @markkastoun
    btw, tried it on Docker for Windows and it works like a charm
    i'll dig more into my server setup
    Pospielov Serhii
    @pospielov
    Probably logs will be too heavy for Pastebin, you can send them to our email: compreface.support@exadel.com
    markkastoun
    @markkastoun
    will do .. any specific logs you require from the containers?
    Pospielov Serhii
    @pospielov
    probably it's better to send all logs you can get. And also information about your operation system.
    Conrado García Berrotarán
    @conradogarciaberrotaran
    Hi, congrats on everyone involved in this project. I find it really cool.
    Does anybody have a real-time project?
    Pospielov Serhii
    @pospielov
    Hi @conradogarciaberrotaran one of our users shared his Frigate integration:
    https://github.com/exadel-inc/CompreFace/discussions/419
    I believe it's real time
    Conrado García Berrotarán
    @conradogarciaberrotaran
    thanks! i'll check it out
    Conrado García Berrotarán
    @conradogarciaberrotaran
    is compreface free and open source forever?
    Pospielov Serhii
    @pospielov
    Look, I can't guaranty anything as I don't own the repository.
    But Exadel mostly is a service and not a product company and the current strategy is "to use a typical open-source monetization strategy, which is a free product with paid support and services. The offered services include integration, customization, new feature development, training, etc".
    I believe that in the nearest future Exadel won't change its strategy
    Conrado García Berrotarán
    @conradogarciaberrotaran
    I asked because I got an exception from running the docker-compose
    something like "DemoNotAvailableAnymore"
    i thought that this was a limited time demo or something, I couldn't find it anywhere on the docs
    Pospielov Serhii
    @pospielov

    oh, probably we need to hide this from log.
    When you start CompreFace the first time, on the login page, there is a "try demo" button.
    So you don't need even to register to try CompreFace
    When you register, there is a "demo" application with the same functionality
    But we hide the "try demo" button for security reasons - only registered users will see it.
    And this logs says that there is no more "try demo" button

    And yes, CompreFace is totally free, with no paid features or restrictions, it could work without the internet and we can't disable your installed copy remotely :)

    Conrado García Berrotarán
    @conradogarciaberrotaran
    i understood, thanks for your responses
    keep up with the great work
    David Jakowenko
    @jakowenko_twitter
    Is there any harm in training the same image multiple images using the API? If I have a container that trains compreface with images from a folder every time it starts, is there any problems if that image has already been sent to compreface?
    Love the project, thank you for the amazing work!
    Pospielov Serhii
    @pospielov
    If you send the same image twice, we record it as two examples twice. Then when we recognize the unknown photo we compare the photo with all of the examples. So as the examples will grow CompreFace will require more RAM and CPU.
    For 50,000 examples it will be about 0.2-second delay in the calculation and about +4Gb of required RAM
    David Jakowenko
    @jakowenko_twitter
    Thanks for letting me know. I'll make sure to manage if that file had already been sent as an example on my end, so I don't retrain it with the same ones over and over.
    I built some routes into my container that pull images directly from frigate to then label and feed into compreface. it makes adding new photos a breeze
    herryzhang
    @herryzhang
    CompareFace can count how many faces are in the video?
    Pospielov Serhii
    @pospielov
    CompreFace can be used for just face detection (and counting of course), but you need to parse video to frames by yourself and send it to CompreFace.
    herryzhang
    @herryzhang
    If there are multiple face images of a person, which one should I use? What should I do with it
    Pospielov Serhii
    @pospielov
    I believe the best is to use the photo with the best quality. But if you have more than one photo - just upload them all with the same name. During recognition, the face recognition algorithm will look through all of them and return the best match.
    Still, I do not recommend uploading faces of bad quality as an example, it could answer then for all bad quality images that this is the same person.
    I did some experiments to prove it:
    https://github.com/exadel-inc/CompreFace/discussions/419#discussioncomment-435968
    sunpeng.fhc
    @sunpeng-fhc
    How much concurrency can a compareface stand-alone support?
    Pospielov Serhii
    @pospielov
    It really depends on the hardware and the images - image size, resolution, and number of faces on it.
    E.g. we tested on images that are 512Kb, 1600x900, one face, CPU with 8 cores - with 8 concurrent requests the response time was less than a second.
    Shiddu
    @shiddugmail
    hello friends I am new to this group, I am working on setting up face enrolment system for academic project. i would like to know how to register new faces in to the system. please help i have installed the 0.5.0.
    Pospielov Serhii
    @pospielov
    Hi, the good way to start is from our "Hor to use CompreFace" guide - https://github.com/exadel-inc/CompreFace/blob/0.5.0/docs/How-to-Use-CompreFace.md
    I added two demos, hope they will be useful
    And if you use JavaScript, you can use CompreFace JavaScript SDK to add new faces: https://github.com/exadel-inc/compreface-javascript-sdk#add-an-example-of-a-subject
    Siva saikumar reddy k
    @sivasai2207
    Can someone help me with this issue..?

    I have tried to train a model by passing a relative path than in that case it works. But to automate the process using the Node Js File system module. It generates the absolute path so I need to pass the absolute path instead of the relative path.

    I have tried implementing an absolute path instead of a relative path. I ran into an error.https://github.com/exadel-inc/compreface-javascript-sdk/discussions/12

    Pospielov Serhii
    @pospielov
    @/all Hi all, we just released a new version of CompreFace - 0.5.1.
    Two main changes:
    1. Added support base64 format for images.
    2. Fixed "Verify Faces from a Given Image" endpoint
      See more information on the release page: https://github.com/exadel-inc/CompreFace/releases
    arladmin
    @arladmin
    Does it also work with people wearing face masks?
    Pospielov Serhii
    @pospielov
    Yes, but the accuracy is not so good as without masks, we use models that were built on datasets with faces without masks.
    Subjectively - the accuracy is still quite good, but the problem is there is no good benchmark to check it.
    Sergey Rodionov
    @sergey64_gitlab
    Do all models available in the CompreFace have no restrictions for commercial use? I've contacted the authors of insightface and they said that their models were trained on datasets which are for non-commercial usage only.
    Pospielov Serhii
    @pospielov

    First of all, I'm not a legal expert, so I can be wrong in any of my claims.
    We do not own the models, just use other open-source libraries, so the licenses of other libraries apply in this case. Both FaceNet and InsightFace are licensed by MIT License which is totally compatible with CompreFace Apache-2.0 License and can be used in commercial use without limitations.
    But if Insightface answered you that it is prohibited to use their models, then probably this is it. When we started to integrate with InsightFace there were no limitations, but as far as I know, they can change the license, and now there are lots of mentions "Only for research purposes" on both their GitHub and pypi.org project.

    The problem is that it's not clear if anybody can use any face recognition model at all(even for non-commercial usage).
    Think about it - to train a face recognition model, you should collect a huge dataset of faces. To use it, even for non-commercial usage, you should receive the consent of every person in this dataset. And this is impossible.
    Just an example - MS-Celeb-1M dataset - which is used by Insightface to train models that are now used by CompreFace.
    https://exposing.ai/msceleb/
    According to this article, Microsoft didn't ask for the consent of any of the people in the dataset. Also, even if it was created for "non-commercial research purpose only", there is information that it was used by commercial face recognition services like Face++, SenseTime, and Microsoft. It was also used by other companies like Canon, Hitachi, IBM.

    So here are my thoughts (which shouldn't be considered as advice for actions):

    1. We really need legal regulations for all of this. So everybody understands the rules.
    2. It's better to use anonymized datasets for training (e.g. MS-Celeb-1M doesn't contain names, but LFW does)
    3. It's more acceptable to use a pre-trained model in commercial usage than using datasets. As commercial companies do not have access to images of faces in this case.
    4. Big companies and governments (including the USA and China) use face recognition models based on datasets created without any consent of people. Are you allowed to do it? Who knows, see my first thought.
    Marc Fielding
    @marcfielding1

    @pospielov hi i've just joined this channel with a question but I can shed some light on this - people in the celeb dataset willing posted those images online for public consumption and thus gave consent - I don't know where you would stand on commercial use of someone's face within a dataset to improve a model, but those images were willingly posted online with no license attached.

    If someone were to walk into an office and slurp their face and apply a label to it(their name) then that's not good. It's why everywhere has "Smile you're on CCTV" signs because people can make a choice if they'd like to be on CCTV or not.

    Also if an agent of celeb etc came to you and said "You're using Pamela andersons face"(first time anyone's cared about her face) - what I would answer is that - yes we have, do you know you can now use this model to detect say deepfake porn or images of Pamela anywhere - so whilst it's very unlikely that someone would complain I'd frame the positives.

    The law does however vary from country to country, with celebs I don't think you're on dodgy ground since you're not making money from their face your using an image they willingly provided to the public.

    Marc Fielding
    @marcfielding1
    Worst thing you'll have to do is remove someone from the detection capabilities i reckon, although i'm not a legal expert in this area(mine is mainly GDPR in the EU) I can't see anyone kicking off that your software recognises them, you're not selling it off the back of you can recognise Pamela, you're doing a demo using images that are freely available.
    Marc Fielding
    @marcfielding1
    articles like that, kinda make me annoyed, I mean their complaining on behalf of someone that's never complained, that posted their face online, anyone that moans about posting online and then someone using what they posted doesn't understand the internet and should take all their devices out the back and smash em.
    Marc Fielding
    @marcfielding1
    Anyway back to my question, I'm trying to run these containers on NVIDIA Jetson's devices such as Xavier and or Nano for GPU's - are you using Buildx etc to support linux/arm,linux/arm64 arch, I'm getting some funky errors.
    Annddd... I think I just found the Dockerfile which I missed because I was too busy playing with your product, ignore my question unless I come back :laughing:
    seawolfer
    @seawolfer
    Hi, new to this :-(
    i have installed all, and it seeemed to work, but, there's NO link to create a new application :-(. Any help?
    Pospielov Serhii
    @pospielov
    Hi, any chance that you encountered this bug? exadel-inc/CompreFace#588
    If yes - please clean the installation and create only one user. He will be an owner and he will be able to create applications and give permissions to create applications to other users