Skip to content

Latest commit

 

History

History
45 lines (36 loc) · 3.75 KB

README.md

File metadata and controls

45 lines (36 loc) · 3.75 KB
Raw

Object-Detection-Projects

This repo contains all the projects related to object detection, OpenCV, MediaPipe, Computer Vision, LabelMe, ALbumentation, VGG16, SSD architectures

1. Face detection Using OpenCV, LabelMe, Albumentation, TensorFlow, Keras, VGG16

  • We started with getting Images using OpenCV
  • Then we labeled the images using LabelMe i-e (bounding box around the face)
  • Then we Augmented those images and labels using Albumentation
  • Then we imported the VGG16 architechute from tensorflow.keras.applications and did not include the top layer
  • Then I added the top layer with my own configuration using Functional API and compiled and trained the model on the augmented data with Total params: 16,826,181
  • Model was saved in h5 file "facetracker.h5"

How to run the working project?

  • You need to run !pip install labelme tensorflow tensorflow-gpu opencv-python matplotlib albumentations os time uuid json at the beginning if these libraries are not installed
  • If you want to see the project working, just run the 04-Real-Time-Detection.ipynb notebook
  • a pop up window will open and move around to see the face detection working
  • To get out from the loop press "q" key on keyboard

Demo

Face Detection

2. Sign Language Detection Using OpenCV, MediaPipe Holistic and LSTM Model

  • We used Open CV to get the images and MediaPipe Holistics for landmark detection
  • Then we extracted keypoints for "pose", "face", "left hand", and "right hand"
  • Then we set up folder to save our data
  • Then we captured 30 images for "Thank you" and labelled them and save them
  • Then we captured 30 images for "hello" and labelled them and save them
  • Then we captured 30 images for "I Love You" and labelled them and save them
  • Then we preprocess the data and created features and target for our training
  • Then we split our data using train_test_split
  • Then we used Sequential API (LSTM) to create our model with Total params: 596,675
  • Then we saved our model as "action.h5"

How to run the working project?

  • You need to run !pip install tensorflow tensorflow-gpu opencv-python matplotlib mediapipe sklearn numpy os time at the beginning if these libraries are not installed
  • If you want to see the project working, just run the 02-Test-in-Real-Time.ipynb notebook
  • a pop up window will open and do the signs in the gif below
  • To get out from the loop press "q" key on keyboard

Demo

Hello Thank You I Love You
Hello Thank You I Love You