ELE510 Project - Hand gesture detection

HaGRID - HAnd Gesture Recognition Image Dataset

Website: https://www.kaggle.com/datasets/kapitanov/hagrid
Github: https://github.com/hukenovs/hagrid

Since the full dataset is 716GB we use the some subsample sets.
Train and validate sets: 500 Images for each class from HaGRID test set where used for training and validating the model
Train images download: https://sc.link/zlGy
Train annotation download: https://sc.link/DE5K

Test set: HaGRID subsambple set where used for testing our model
Test images download: https://sc.link/AO5l
Test annotation download: https://sc.link/EQ5g

Folder structure

Since the combined image files are so large we could not add them to the repository. The only folders that has to be added are under images.

├───manual_test
│   ├───grayscale
│   └───rgb
├───test_original
│   ├───call
│   ├───dislike
│   ├───fist
│   ├───four
│   ├───like
│   ├───mute
│   ├───ok
│   ├───one
│   ├───palm
│   ├───peace
│   ├───peace_inverted
│   ├───rock
│   ├───stop
│   ├───stop_inverted
│   ├───three
│   ├───three2
│   ├───two_up
│   └───two_up_inverted
├───test_preprocessed
│   ├───call
│   ├───dislike
│   ├───fist
│   ├───four
│   ├───like
│   ├───mute
│   ├───ok
│   ├───one
│   ├───palm
│   ├───peace
│   ├───peace_inverted
│   ├───rock
│   ├───stop
│   ├───stop_inverted
│   ├───three
│   ├───three2
│   ├───two_up
│   └───two_up_inverted
├───train_original
│   ├───call
│   ├───dislike
│   ├───fist
│   ├───four
│   ├───like
│   ├───mute
│   ├───ok
│   ├───one
│   ├───palm
│   ├───peace
│   ├───peace_inverted
│   ├───rock
│   ├───stop
│   ├───stop_inverted
│   ├───three
│   ├───three2
│   ├───two_up
│   └───two_up_inverted
└───train_preprocessed
    ├───call
    ├───dislike
    ├───fist
    ├───four
    ├───like
    ├───mute
    ├───ok
    ├───one
    ├───palm
    ├───peace
    ├───peace_inverted
    ├───rock
    ├───stop
    ├───stop_inverted
    ├───three
    ├───three2
    ├───two_up
    └───two_up_inverted

Train_original and test_original contain the images used for training/validating and testing the model. Train_preprocessed and test_preprocessed contain the preprocessed images used by the model. The preprocessing is done in the notebook.