Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
  • Mar 19 22:11

    dependabot[bot] on pip

    Bump jinja2 from 2.7.3 to 2.11.… (compare)

  • Mar 19 22:11
    dependabot[bot] labeled #38
  • Mar 19 22:11
    dependabot[bot] opened #38
  • Aug 04 2020 02:44
    vishalsingha opened #37
  • Nov 04 2019 00:26

    dependabot[bot] on pip

    Bump requests from 2.5.1 to 2.2… (compare)

  • Nov 04 2019 00:26
    dependabot[bot] labeled #36
  • Nov 04 2019 00:26
    dependabot[bot] opened #36
  • Sep 27 2017 13:54

    dexter1691 on master

    Markdown in readme Fixed Update README.md Merge pull request #4 from Harg… (compare)

  • Sep 27 2017 13:54
    dexter1691 closed #4
  • Sep 27 2017 13:54
    dexter1691 closed #3
  • Sep 27 2017 13:54
    dexter1691 synchronize #4
  • Sep 27 2017 05:52
    HargovindArora commented #4
  • Sep 18 2017 16:37
    HargovindArora commented #3
  • Sep 18 2017 16:35
    HargovindArora opened #4
  • Sep 18 2017 14:16
    HargovindArora edited #3
  • Sep 18 2017 14:15
    HargovindArora opened #3
  • Nov 27 2016 20:26
    anweshknayak commented #22
  • Sep 16 2016 17:16
    prateekarora151083 opened #35
  • Dec 08 2015 04:25
    katienotkathy opened #34
  • Dec 05 2015 01:03

    dexter1691 on master

    Update README.md (compare)

Harsh Agrawal
@dexter1691
The task is somewhat tangential to the actual goals/ tasks of GSOC so I highly recommend writing a decent proposal for those ideas.
and its a 20 minute long video! can you create like a summarized version of it :P
Surya Teja Cheedella
@SnShine
@dexter1691 Yeah, Thank you. And the first 10 minutes is briefly explaining the code, you can view it from 10:20 to see the code running and printing output accuracy.
Harsh Agrawal
@dexter1691
sounds good. :)
Suvil
@suvil
@dexter1691 - I missed that part. My apologies. It wouldn't distribute tasks as it is without knowing the target machines capabilities. It was more of a design thing - to combine the two tasks. That there is only one interface to control them. I find it kind of redundant to see a load balancer distributing work b/w IaaS machines while another system distributes tasks to consumer devices. I was hoping to make them more homogenous. The system would be intelligent enough to determine and discriminate b/w the processing power of different devices but this would appear seamless to people using the system.
@dexter1691 - also, has the cloudcv team looked into BOINC? It's used by SETI for its SETI@home project to achieve distributed computation on consumer devices.
Harsh Agrawal
@dexter1691
oh I see what you are saying. yeah, you are right. It will still require different efforts because VMs are kind of more dedicated and will reside on a single network so its gonna be better for parallel jobs. For consumer devices, we will have stronger consistency and will have to take care of multiple failure causes.
And yes, the inspiration came from BOINC like projects.
Suvil
@suvil
Yes. Those are caveats that we have to keep on mind.
Any idea how high this would be on the priority list?
I am seriously interested in writing a proposal.
Harsh Agrawal
@dexter1691
As I have said before, it all depends on the number of slots we get and how good / realistic the proposals are.
Smriti Jha
@smritijha
Hey @dexter1691, I just uploaded two random images to stitch here ( http://cloudcv.org/image-stitch/ ), and I see them being run and saved on terminal, but I don't see the final panorama.
Harsh Agrawal
@dexter1691
Hmm... interesting. Does the terminal say it got finished. It might have got stuck somewhere. Its a bug in a code that was written by Dhruv's student.
I think I fixed it but seems like we are using the wrong executable.
I will get in touch with prakriti to check on this. Thanks
Smriti Jha
@smritijha
Well, I ran it again, but with the images you have provided. I downloaded and then uploaded them though. That runs correctly. Terminal closes with saving all image files (ex. Saving file:P1020358.jpg) as it did previously too.
Harsh Agrawal
@dexter1691
yeah. Seems like our code is not getting the error stream to show the error on the terminal.
if you run sample images, or images that stitch well then there the server is able to find the image and returns that.
Smriti Jha
@smritijha
Yes, I figured. So, I refreshed the page, ran the original two images I was running, copied the terminal output, compared it with the output from the successful run with sample images (https://www.diffchecker.com/xlb9ba4b). My two images were different in sizes though. So, I downloaded a Panorama image from the interwebz, split it into 4 images and ran it on CloudCV. All splits being equal in dimensions, just like the sample images were. That did not work either. The sample images are 500 x 334 and the panorama images I am using are 2394 x 600.
Smriti Jha
@smritijha
Resizing hasn't helped either.
Harsh Agrawal
@dexter1691
@/all Those who are working on submitting a proposal for Dropbox integration / Google Drive integration should look into how can we access various photo collections (dropbox, facebook, google+, flikr) for our "Very Important Person" Demo.
The idea is to sort the photo collection where you turn out to be the most important person in the image. We can also think about a more researchy side of it where we also take into account the context, scene recognition etc!
I was hoping for you people to think about such an application on your own but seems like a lot of people didn't think about it.
Bharat Gaind
@bharatgaind234
Hey everyone,
I've been working on the web-d demo app and I'm almost done. Could anyone please tell me where I can find the caffe API, which is supposed to train the classifier and generate the results? Or should I just create a video of everything other than that and send it?
Also, what is the deadline for this demo-app submission? Thanks.
Anuj8826
@Anuj8826
hey , i wanted to tell that there are some tricks through which we can access collection of images by directly putting in query an image model of that particular class
smahanthi
@smahanthi
Hey dexter, can you take a look at my video https://www.youtube.com/watch?v=UntUciMJpeA and tell me what else I need to do to complete the requirement for proposal?
Thanks.
Haider Adeel Agha
@haideragha123
Hi dexter, regarding the cnn task, if I have not achieved reasonable results should i continue doing it?
Utkarsh Sinha
@liquidmetal
Hey @dexter1691, can you take a look at http://cloudcv.utkarshsinha.com/ for the webdev task? Any suggestions?
Harsh Agrawal
@dexter1691
@liquidmetal : Looks pretty neat. I think, when I press train, it just trains for one category.
I might be wrong
@haideragha123 : It's fine. The application deadline is over. I hope you were able to tell us whatever you were able to do in the proposal.
Utkarsh Sinha
@liquidmetal
@dexter1691 Check the "category" on the job details page - I tried a few different categories and it seems to be training different ones.
You need to create multiple categories first, though.
Harsh Agrawal
@dexter1691
yeah. I guess, but it sure is a good interface.
Utkarsh Sinha
@liquidmetal
@dexter1691 thanks!
Krishna Bagadia
@krishna95
Hello Everyone!! I was wondering regarding the "Very Important Person" problem on cloucv website. if there is only one person in a single image, then should that person be treated as a VIP?.
Thanks in Advance!!
Harsh Agrawal
@dexter1691
@krishna95 : Well why not? if he is the only person in the image, he is still important. Think about its applications, if say you are sorting albums based on importance of a particular person , then shouldn't this image rank higher than other images where he is not the most important person in the image?
Krishna Bagadia
@krishna95
@dexter1691 :smile: I totally agree with you. But when I ran the function on images containing only one person(like this), it showed "Oops! We could not detect any face in this image.This demo works best if there are atleast two people in the image, as found by a face detector. Please try another image." Is there anyway where I could help in this regard?
Harsh Agrawal
@dexter1691
Amm.. so the face detector failed to detect a photo which is why it says that try a new image.
Krishna Bagadia
@krishna95
I read the paper regarding the same and they are using the skybiometry for the face detection. I uploaded the same image there, but it detected the face correctly.
Harsh Agrawal
@dexter1691
@mclint : Do you want to answer this? @krishna95 claims that skybiometry was able to detect faces on an image, but your system doesn't :P
Harsh Agrawal
@dexter1691
BTW Clint mentioned that he is not using skybiometry for the demo. I have asked him to explain here in more detail.
mclint
@mclint
Hey Guys, the VIP demo does not use skybiometry API due to licensing constraints. It just uses the opencv face detector. Also, the model used in the Demo for prediction is not the same as in the paper. The paper uses DPM pose estimation and other computationally expensive features. We wanted to keep the demo simple and fast.
As far why we do not consider a single person to be important, that is a trivial case which can be considered as important or not based on the application that eventually runs on top of importance. There is no point in asking a question - Who is the most important person? - when there is just one person. The more interesting question would be, in which image is this person more important - this is also part of the paper - a.k.a the corpus level task.
Deshraj Yadav
@deshraj
This message was deleted
Shubham Desale
@shubhamd
Hello everyone! I was working on documenting the current PCloudCV package and have setup a basic structure using Sphinx here : batra-mlp-lab.github.io/pcloudcv . I am finishing up the class level documentation, any suggestions/ideas would be welcome! :smile:
Deshraj Yadav
@deshraj
Hey @shubhamd, great work. :)
Shubham Desale
@shubhamd
Thanks @DESHRAJ .
Krishna Bagadia
@krishna95
@mclint and @dexter1691 : Thanks for the clarification :).