underwater-gesture-recognition

This repository contains the codes used in the paper "Underwater Gesture Recognition Using Classical Computer Vision and Deep Learning Techniques". The codes are divided into three groups: one for each model used in the paper.

1. Bag of Visual Words (BOVW):

• Scripts (adopted from Kushal Vyas’ implementation of Bag of Visual Words: https://github.com/kushalvyas/Bag-of-Visual-Words-Python)

• To test:

  1. Put train images in ./images/train folder; this is needed to get the classes to be used in testing
  2. Put test images in ./images/test folder
  3. Put kmeans_cluster_3.sav, mega_histogram_3.pkl, svm_train_3.pkl, and vstack_3.pkl in the same directory as Bag.py and helpers.py (Link to pretrained weights: https://drive.google.com/open?id=1hSaZwRpbtOqFYep7Z4jW2kkKizEHoVqO) 4. Run python Bag.py

• Dependencies:
  -OpenCV 3.4.2.17 (OpenCV-contrib-python==3.4.2.17), scikit-learn 0.20.3

2. Histogram of Gradients (HOG):

• To test:
  1. Put test images in ./images/test folder 2. Put hog_svm.joblib pre-trained model in ./models folder (Link to pretrained weights: https://drive.google.com/open?id=1pirGWIkZqWXBNSwQKuTdLpmMKowYrOUK)
  2. Create ./results/ folder, where the csv containing the target classes and predictions will be saved
  3. Run testing.py
• To visualize sample correct and incorrect predictions per class:
  1. Create ./visualization/ folder, which will contain the sample images with predictions
  2. Run visualize_predictions.py
• Dependencies:
  -OpenCV 4.1.0.25 (opencv-contrib-python==4.1.0.25), scikit-learn 0.21.0

3. ResNet50-CNN

• To test:
  1. Place the test in ./images/test folder
  2. Put the pre-trained model resnet50_CNN_caddy.pickle in the same directory as classifier_withtest.py (https://drive.google.com/file/d/1EQxRwaFTZVlmVlTZDVx5Y1fcLAxjL6Qj/view?usp=sharing)
  3. Run classifier_withtest.py
• Dependencies:
  • pytorch 0.4.0
  • cuda 9.0

Datasets

• Link to train images: https://drive.google.com/open?id=1JCXaT9PSnyJ4KgTzlumMrmwALVt-2V_t
• Link to test images: https://drive.google.com/open?id=1ducpYZnC9dQznG90HbiyZSX3CYCR8UbR