Xuanhao

Robotic Arm to Automatically Changing 3D Printer Plates

Sun, 01 Dec 2024 16:40:57 -0400

Adversarial Video Generation

Design and implement a robotic arm system capable of automatically removing finished 3D print plates (for Bambulab A1C) and replacing them with new ones.
Implement ROS nodes for motion planning, sensor data processing, and error handling, ensuring robust and reliable system operation.
Started from self-implemented inverse kinematics algorithms to leverage MoveIt! within ROS to generate smooth and collision-free trajectories for the robotic arm. Working on integrating LLM APIs to the system.

API tools for Cryptocurrency quantitative analysis

Mon, 01 Feb 2021 16:40:57 -0400

Quantitative analysis tool based on the CCXT, which is a JavaScript / Python / PHP library for cryptocurrency trading and e-commerce with support for many bitcoin/ether/altcoin exchange markets and merchant APIs.

Features

Ping all active training pairs (e.g. BTC/ETH, BTC/USDT etc.) and cache for later usage.
Fetch all active training pairs price difference in one specific exchange from STARTTIME to ENDTIME in Timestamp.
From public API, no requirement for API key or registration.
Export to *.xlsx

Demo

To-do

Add an argparser
Multiprocess support
GUI and k-line

Movie Recommendation System

Thu, 01 Oct 2020 17:30:40 -0400

Movie Recommendation System

Built a movie recommendation system based on Collaborative Filter in PySpark on the MovieLens Dataset.
Conducted ETL scripts to ﬁlter movies’ categories, rating or other information with Pandas and Spark.
Analyzed and compared the performance with other cluster algorithms like K-means and DBSCAN.

Web-based Trash Classifier

Wed, 01 Jan 2020 17:30:40 -0400

Web-based Trash Classifier

Designed a web-based trash classifier using Javascript and TensorFlow.JS. Users can get access through both mobile phone or website.

YelpBBQ

Wed, 02 Oct 2019 16:32:52 -0400

YelpBBQ

A RESTful Node.js web application project that allows users to share their BBQ experience by uploading photos and comments.

Live Demo

To see the app in action, go to https://calm-depths-02705.herokuapp.com/

Username: guest Password: guest

Features

Authentication:
- User login with username and password
- Admin sign-up with admin code
Authorization:
- One cannot manage posts and view user profile without being authenticated
- One cannot edit or delete posts and comments created by other users
- Admin can manage all posts and comments
Manage campground posts with basic functionalities:
- Create, edit and delete posts and comments
- Upload campground photos
- Display campground location on Google Maps
- Search existing campgrounds
Manage user account with basic functionalities: Password reset via email confirmation~~ (disabled)
- Profile page setup with sign-up
Flash messages responding to users’ interaction with the app
Responsive web design
Custom Enhancements
Update campground photos when editing campgrounds
Update personal information on profile page
Improve image load time on the landing page using Cloudinary
Use Helmet to strengthen security

Front-end

Back-end

Platforms

What Next

Immigrate to AWS and DynamoDB

Google Analytics Customer Revenue Prediction

Tue, 01 Oct 2019 17:30:40 -0400

Analyzed raw log data from GStore (faced to normal customers) to predict revenue per customer.
Preprocessed the raw data (2.4GB) based on Pandas and Numpy and made data visualization based on Matplotlib and Seaborn module.
Improved the data procession speed based on parallelization using Hadoop and PySpark.
Trained model to predict based on LightGBM, XGBoost and CatBoost, assembled these models to improve 20% performance.

Adversarial Video Generation

Tue, 01 Oct 2019 16:40:57 -0400

Adversarial Video Generation

This project was forked from dyelax’s work and immigrated to TensorFlow1.2 and Python3, in order to be compatible to GTX1080 or later version GPU.

Adversarial generation uses two networks – a generator and a discriminator – to improve the sharpness of generated images. Given the past four frames of video, the generator learns to generate accurate predictions for the next frame. Given either a generated or a real-world image, the discriminator learns to correctly classify between generated and real. The two networks “compete,” with the generator attempting to fool the discriminator into classifying its output as real. This forces the generator to create frames that are very similar to what real frames in the domain might look like.

Results

We collected the data from a basketball game and a video game Snake.

After the first 10000 times training, the last frame of the Snake Game looked still fuzzy.

The detailed results is shown at here.

After another 40000 times training, the basketball sence looked like still fuzzy, while the Snake sence looked better.

An interesting finding was that when the snake met the cherry, which was used to earn points, it felt “puzzled” about which direction to take. This “puzzled” also exists when the snake went close to the wall or its own body.

Pictures Histogram Accelerating based on CUDA

Mon, 01 Oct 2018 16:40:57 -0400

Pictures Histogram Accelerating based on CUDA

Implemented a histogram equalization algorithm based on CUDA, where the benchmark of GPU was Nvidia Kepler 10. To optimize, I downsampled 16 pixels (144 bits) as a block to the kernel at a time, which was a tradeoff between accuracy and efficiency. Also, I pinned-memory on the host side and optimized the blocks with a one-dimensional block design.

Reinforcement Learning in Game Scenarios

Fri, 01 Sep 2017 16:40:57 -0400

Reinforcement Learning in Game Scenarios

Implemented agents to play games automatically based on reinforcement learning (Q-learning). The game scenarios include a basic Taxi game, Flappy Birds and an Atari Game: Space Invader.

Interesting Result

Dublin

Fri, 01 Jan 2016 16:40:57 -0400

Dublin

Dublin is a legendary story of a hero who fighting against the dragons and demons to rescue his love.

I implemented the Java GUI application based on Swing Framework. The most interesting part was I hand-drawed about 20 8-bits pictures as the characters, environments and actions.

The Demo may be found HERE.

Xuanhao

Robotic Arm to Automatically Changing 3D Printer Plates

Adversarial Video Generation

API tools for Cryptocurrency quantitative analysis

Features

Demo

To-do

Movie Recommendation System

Movie Recommendation System

Web-based Trash Classifier

Web-based Trash Classifier

YelpBBQ

YelpBBQ

Live Demo

Features

Custom Enhancements

Front-end

Back-end

Platforms

What Next

Google Analytics Customer Revenue Prediction

Adversarial Video Generation

Adversarial Video Generation

Results

Pictures Histogram Accelerating based on CUDA

Pictures Histogram Accelerating based on CUDA

Reinforcement Learning in Game Scenarios

Reinforcement Learning in Game Scenarios

Interesting Result

Dublin

Dublin