Xinjia Pang

ML Tools

 

 

RUNWAY ML

BIGBIGAN Model

This model in Runway Ml generates images using the BiGAN Encoder based on BigGAN. We used these photos we took from the Art Gallery and put them into the models. This BigGAN comes out several results are interesting.

Austin Thomas’s Work

Import

Import

Export

Export

Danielle Dimston’s Work

Import

Import

 
Export

Export

Yvette Taminiau’s Work

Import

Import

 
Export

Export

Bara Jichova Tyson’s Work

Import

Import

 
Export

Export

 
 

FACE-RECOGNITION model

This model uses the Face Recognition Python library. It can do two commands that allow people to either detect multiple faces in a single input image or identify a specific face using one input image and one label image. I input one image and use the webcam as another input. The model is not that good on identification of face. when my friend showed up on the webcam, the model still labels her like me.

The algorithm for the models is the LBPH Algorithm which is a simple yet very efficient texture operator that labels the pixels of an image by thresholding the neighborhood of each pixel and considers the result as a binary number.

 

YOLACT model

Real-time Instance Segmentation

Uses ResNet-101 with FPN (Feature Pyramid Networks) helps in creating pyramids of feature maps of high-resolution images rather than the Conventional Pyramid of Images approach, therefore faster than the other competitive frameworks

Insufficient:

The program is trained to identify several types of plants. The database is limited and cannot support the identification of a large number of varieties.

WEKINATOR

In project 2, I create 3 different examples of the Wekinator classification expansion pack. I put several two-color dots on the frame, the classification will identify the dots different based on the algorithm I changed.

 

K-Nearest Neighbor

K-Nearest Neighbor.jpg
Naive%2BBayes.jpg

Naive Bayes

Decision Tree

Decision Tree.jpg
AdaBoost.M1.jpg

AdaBoost.M1

ML5.JS/P5.JS

p5.png

ML Online Edit

Using MobileNet architecture (proposed by Google) to analyze the images classier.

The database is used to train ML is from Imagenet. The list in the link shows the items can be identified by the program:

https://github.com/ml5js/ml5-library/blob/master/src/utils/IMAGENET_CLASSES.js

Code: https://github.com/CodingTrain/website/blob/master/Courses/beginner_ml5/03_video_classification/sketch.js

Insufficient: Not accurate enough to easily identify other objects around and cause errors solution in the results. The program is trained to identify several types of plants. The database is limited and cannot support the identification of a large number of varieties.

IMAGE PROJECT IN TEACHABLE MACHINE

Web tool to create machine learning models (no coding required)

Train a computer to recognize your images, then export your model for sites, apps, and more.

Proposed by Google Creative Lab

The models make with Teachable Machine are real Tensorflow.js models that work anywhere javascript runs, so they play nice with tools like Glitch, P5.js, Node.js & more.

Insufficient:

Training by yourself(but easy)

https://teachablemachine.withgoogle.com/models/IYhwSB6P/