by

Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Han Xiao
    @hanxiao
    test
    dongzhuoyao
    @dongzhuoyao
    饭特稀
    Huang Tao
    @mwsht
    Who have the accuracy of WRN/ResNet18 without pre-processing
    雾雨魔理沙
    @MarisaKirisame
    Hey, had anyone have accuracy of DenseNet-BC 768K params without preprocessing/augmentation?
    刘恒
    @liuheng2cqupt
    hello every body !
    Rohan Relan
    @hanrelan
    Hi, I was wondering if you'd be interested in hosting fashion-mnist on brine.io
    brine makes it really easy for people to explore the dataset (we're adding annotation-level filtering soon) and download and use the dataset with keras/pytorch/tensorflow etc.
    @hanxiao let me know if you're interested and I can set that up!
    Sushrut Thorat
    @novelmartis
    The human performance benchmark provides no links to a resource explaining the method of data acquisition or access to the actual data. Could you create that link?
    Eric Rubiel
    @Rubiel1
    Hey does any of you have NN already trained on the fashion dataset?
    Eric Rubiel
    @Rubiel1
    Tanmoy Chowdhury
    @tanmoy1234
    Hello sir , i just saw the mnist dataset , i want to create dataset for character classification (ol chiki - a tribal language ) how would i create a dataset like this . I have written down some of the lettar in paper and scanned it , what is the next step . Thank you .
    Eric Rubiel
    @Rubiel1
    @tanmoy1234 hey that may be more related to the original MNIST and NIST https://www.nist.gov/srd/nist-special-database-19 and I could help you if that is ok.
    Tanmoy Chowdhury
    @tanmoy1234
    In nist , it is english language i think ,but i want to create about other langugae one of them is ol chiki . How did you create such dataaset . What is the algorithmic procedure
    Eric Rubiel
    @Rubiel1
    I didn't created it but I have done ML for years by now, drop me a line: eric.rubiel(at)u.northwestern.edu
    In general you will need a lot of people to write the characters
    Then you scan those characters (you need to control the quality up to some degree, like make sure to separate those characters)
    Then you fix a dimension, and create images of that fixed dimension, ideally the characters should be centered
    a big part of this can be automatized but you need to plan ahead the process, you also need to keep track of the label: for every image, what is the character associated to the image.
    OliverKanat
    @OliverKanat
    Hey all, I'm trying to preprocess images of clothing from another fashion store into mnist format and then trying to get embeddings from them. Shifting them into grayscale and resizing to 28x28 seems fine, it's just they all have white background. Changing the background makes black clothing blend in with the background. Any ideas what would you do?
    Eric Rubiel
    @Rubiel1
    you dont want the background to be noise that distracts algorithms, does a white background gives you a problem with white clothes?
    siham1996
    @siham1996
    can anyone give me a link for that
    i want to download fashion-mnist dataset by images
    OliverKanat
    @OliverKanat
    OliverKanat
    @OliverKanat
    @Rubiel1 In the end I made image mask black and added little bit of white to the rest of the picture, so there would be some difference between background and black clothing. I havent tried using white background, but this approach seemed to work OK
    siham1996
    @siham1996
    @OliverKanat thank you so much
    siham1996
    @siham1996
    i have error while executing https://stackoverflow.com/questions/55049511/how-to-download-mnist-images-as-pngs which is FileNotFoundError: [Errno 2] No such file or directory: '-f/train-labels-idx1-ubyte'
    sayam sawai
    @sayamss
    hello guys
    siham1996
    @siham1996
    @sayamss hi