Our Projects
Releasing my Face detection Convenience library called Bleedfacedetector
” Its a python package that lets the user use 4 different types of face detectors by just changing a single line of code “
About these face detectors:
1. Haar cascade/ viola Jones based detection
2. Hog (histogram of oriented gradients) based detection/
3. SSD + Resnet10 based detection
4. CNN based detection
If you want to read up on the pros, cons, benchmarks between these methods then read the following excellent blog post on it.
post by ( Vikas Gupta @ learnopencv )
Blog Post Link
Official PyPIlink
Github Link
Created by: Taha Anwar
Pedestrian Detection with Classical Vision, Machine learning & Deep learning
Pedestrian Detection with 4 Different Computer vision techniques Method 1: Background subtraction + Contour Extraction.
Method 2: Haar Cascades (Viola jones classifiers) Method 3: Histogram of oriented gradients and support vector machine. Method 4: Single shot detector + MobilenetNow as you can see the video feed is highly unstabilized but still some techniques are effective, Notice that SSD does not perform better as it should I assume that one of the reasons is the Asian clothing bias of the detector.
Github Link
Created by: Taha Anwar
Hand Gesture Recognition based Calculator (works without ML or DL)
The fact this works without ML and DL means it can run in real-time on something like a raspberry pi.
So how did I do this? last time I created a boring Shape detector, so this blog is built on top of it and now instead of shapes it recognizes basic hand signs.
Github Link
Created by: Taha Anwar
Shape Detection From Pic to Paper without ML or DL
Shape Detection From Pic to Paper without ML or DL. So here I show how we can leverage classical vision techniques to do shape detection on different types of image data streams. So I do shape detection on picture, GUI, and paper here.
Github Link
Created by: Taha Anwar
Automating Games with Dumb AI + Computer Vision
We all know that when it comes to Ai playing Games then Reinforcement learning is the goto strategy but what if the task or game at hand is relatively simple, then we can use some dumb AI (rule-based logic), some GUI automation with some computer vision and viola you have bot playing games. So here I’ve made two algorithms to play some really simple games.
The infamous google dino game:
Some shooting Flash game on scratch:
Github Link
Created by: Taha Anwa
Complete Document Scanner pipeline using Raw Opencv and Numpy
Complete Document Scanner pipeline in just 130 lines of code with raw OpenCV and NumPy Note: I later added Tkinter to let the user select images. Note: I also made an exe of this with pyinstaller but not sharing cause the size was approx 250 Mb. This scanner is part of our Computer vision course in which not only do I explain you every single line of code here but we also develop miniature scanners leading up to this one.
Github Link
Created by: Taha Anwar
Training a Custom Object Detector with DLIB & Making Gesture Controlled Applications
So this is my Hog + SVM based hand palm detector which I trained with dlib on my hands. Instead of manually creating bounding boxes I created an automated sliding window cropper to automate the data collection process.
Github Link
Created by: Taha Anwar
Playing Google Dino Game with Computer Vision
Imagine the ability to play almost any game with gestures and without any special sensors…that’s what computer vision can do for you.
So here in this game, I can jump by opening my mouth and slide by making my face closer.
The best part is that I’ve made this algorithm highly accurate with reasonable fps which is the reason I can easily get so far in the game.
Note: the code for this is not open source right now but I may release it later.
Created by: Taha Anwar
Snapchat like Face Filters using Opencv,dlib & Caffe models
Version 1: SSD+Mobilenet face detecting and masking combined with dlib to account for rotation Version 2: Addition of mouth and eye overlay triggers like Snapchat Note: Opencv is not meant for 3d overlays and it takes a lot of tricks to even make a decent overlay working cleverly you can tell by the fact that there is not much coverage on the topic on the net with regards to Opencv still More improvements and optimizations can be made here.
Github Link
Created by: Taha Anwar