1 |
Face Unmaskers
|
Face Mask Detection and Unmasking the Masked Face using Deep
Learning
|
The trend of wearing masks in public is growing in recent years all
over the world due to COVID-19 epidemic. In particular, public
security check systems can not effectively recognize the masked
faces, but removing masks for passing authentication will increase
the risk of virus infection. We break the problem into two stages:
mask object detection and image completion of the removed mask
region. The results in this project suggest that Face Mask Detection
and cGAN are a promising approach for many Masked Face-Unmasked Face
translation tasks, especially those involving security applications.
Future work includes, adding accuracy metrics, and implement the
system for an embedded system.
|
|
2 |
Hawaii |
Self-Healing FBG Sensor Network For Detection of Bragg Wavelengths
Based on GRU Techniques
|
In this project work, we proposed and designed the reliable
self-healing fiber Bragg grating (FBG) sensor network for improving
FBG sensor capacity in the network based on intensity and wavelength
division multiplexing (IWDM), and effectively detected the central
Bragg wavelength of five FBGs using gate recurrent unit (GRU)
techniques. In this work, we conducted the experiment strain applied
to FBG1 sensor, while the other four FBGs keep fixed. During strain
applied to FBG1, the Bragg wavelength of FBG1 shifts to long
wavelength. In this situation, the spectra of five FBGs are found
either non-overlapped or partially overlapped or completely
overlapped. However, try to identify the Bragg wavelength of each
FBG from partially or completely overlapped spectra using the
traditional peak detection (CPD) method is very challenging and may
not be accurate. To solve this limitation of CPD techniques, we
proposed the GRU algorithm to accurately identify the Bragg
wavelength of FBGs in the condition of the partially or fully
overlapped spectra. The Bragg wavelength detection result shows that
our proposed GRU model have a good performance in terms of accuracy
and computational time to detect accurate Bragg wavelength of each
FBG compare to LSTM and MLP models even the spectra of three FBGs
are overlapped.
|
|
3 |
Musician |
Music Sheet Recognition Base On Deep Learning
|
Brief Introduction of Mask RCNN and model application
|
|
4 |
Text Creators |
Text Generation using Recurrent Neural Network
|
Recurrent Neural Network is natural termed for the great
process in the Natural language processing task. However, in most
previous works, the models are learned based on single-task
supervised objective, which often suffer from insufficient training
data. In this project we trained our model so efficient and
compared with six different models and trained with the high
accuracy model on our own dataset. Experiments on five benchmark
text generator, for instance if you want an articles in high grammar
articles and highly fluent you can use this method to produces
that by giving a name of country it will provide you the historical
moments of that country. The entire network is trained jointly
on this entire task. Furthermore, DL approaches that have been
explored and evaluated in different application domains in NLP
are included in this survey.
|
|
5 |
Lab416 |
Generating Handwritten Number with CGAN
|
Conditional GAN is a variant type of GAN, which is able to decide
type of generated image, to produce the right image we want. We
proposed a number generator model, and using CNN-trained model to
classify generated images, to evaluate our CGAN performance.
|
|
6 |
cocksheeproach
|
breast tumor object detection
|
According to statistics from the Taiwan’s CDC, the top ten causes of
death among citizens from 2018 to 2020 are also cancer. The fourth
leading cause of female death is breast cancer. The order is right
behind to lung cancer, liver cancer, and rectal cancer. They are all
caused by various environmental factors and lifestyle habits in
recent years. Mammography has been regarded by medical experts at
internal and external as a non-invasive detection method for
detecting breast cancer in early stage and effectively reduce the
mortality rate. So, we decided to combine mammography with deep
learning as the topic for improving accuracy of detection. Expect to
reduce the proportion of deaths which caused by breast cancer.
|
|
7 |
MW-LAB |
DeepFaceLab |
DeepFaceLab is a software to change someone’s face in the image or
video, we use it to change the face of video for the final result.
|
|
8 |
SPIE 109 |
Traffic Signs Recognition
|
There are several different types of traffic signs like speed
limits, no entry, traffic signals, turn left or right, children
crossing, no passing of heavy vehicles, etc. Traffic signs
classification is the process of identifying which class a traffic
sign belongs to.
In this Python project example, we will
build a deep neural network model that can classify traffic signs
present in the image into different categories. With this model, we
are able to read and understand traffic signs which are a very
important task for all autonomous vehicles.
|
|
9 |
Deep_Learner |
Deep Residual Network-based Sentiment Analysis of Amazon Cell Phone
Reviews
|
Sentiment Analysis (SA) is a progressive research area in the
process of text mining. SA or the Opinion Mining (OM) is the process
of examining people’s opinions, emotions and attitudes towards the
entity. The entity represents the events, inpreiduals, and topics.
This paper focuses to design a novel classifier approach for the
sentimental classification based on the amazon cell phone reviews.
Initially, the input cell phone review is pre-processed by employing
the stop word removal, and stemming technique. In the feature
extraction phase, the all-caps, emoticon, numerical words, hash
tags, sentiwordnet, elongated words, punctuation marks, and negation
features are extracted in an efficient way from the pre-processed
output. Moreover, the sentimental classification performance is
carried out using the developed Deep Residual Network classifier
approach. Meanwhile, the efficiency of the developed Deep Residual
Network technique is computed based on the performance metrics, such
as accuracy, sensitivity, and specificity. Besides, the proposed
Deep Residual Network approach obtained efficient classification
performance with respect to the maximum accuracy of 0.953, higher
sensitivity of 0.944, and maximal specificity of 0.944,
respectively.
|
|
10 |
Ahpekoy |
Classification of Image Category using CNN in Matlab
|
We use the CNN function of Matlab to build a model to classify the
image of our member's cat
|
|
11 |
Super Mario |
A research on Prognostic and Health Management — Taking Fuse Heating
System as an Example
|
The development of Prognostics and Health Management (PHM)
technology has allowed humans to move from passively solving the
failures of machinery and equipment to proactively predicting when
abnormalities and failures will occur. This technology not only
allows machinery and equipment to operate more reliably, At the same
time, losses are reduced. This research mainly analyzes the characteristics
of the fuse heating system to determine whether the
process is normal or abnormal, and further explores the detailed
classification of abnormal processes, uses random forest to perform
feature selection to reduce the amount of noise, then constructs
with 1D-CNN model. The model is constructed with two sampling
methods (random sampling, stratified sampling) and finally evaluated
by 4 indicators in the confusion matrix, which are accuracy,
precision, recall rate, and F1 score. The result of the study is
that stratified sampling is better than random sampling in all four
indicators of the confusion matrix.
|
|
12 |
too 甜配 cola |
GitHub Project Analysis
|
This is our final project - GitHub Project Analysis, which talking
about CAPTCHA break.
|
|
13 |
WD40 |
Object Detection Classifier for Multiple Objects
|
Image classification involves assigning a class label to an image,
whereas object localization involves drawing a bounding box around
one or more objects in an image. Object detection is more
challenging and combines these two tasks and draws a bounding box
around each object of interest in the image and assigns them a class
label.
|
|
14 |
HSC Lab |
Face Mask Detection Using YOLOv4 Tiny
|
Our project use the latest YOLOv4-tiny as our real time object
detection method and achieved mask recognition
|
|
15 |
Wen’s Team |
Advertisement Spam Detector w/ LSTM
|
Using bidirectional LSTM to train a binary classification model that
can tell if a text contains AD or not. predict model deployment +
Email warning system and parsing tweets from twitter developer api
are build with Flask.
|
|
16 |
深度學習最後專案 |
face mask recognition
|
face mask recognition
|
|
17 |
AI School |
Decision Tree & Confusion Matrix
|
Pump Sensor Data for Predictive Maintenance
|
|
18 |
Speaker Recognition
|
Speaker Recognition
|
Our project is to recognize the speaker of audios. We select MFCCs
to extract the features and use MLP, LSTM, and CNN techniques to
train the model.
|
|
19 |
One Man Team |
Music Generation using Deep Convolutional General Adversarial
Networks
|
Generating music has a few notable differences from generating
images and videos. First, music is an art of time, necessitating a
temporal model. Second, music is usually composed of multiple
instruments/tracks with their own temporal dynamics, but
collectively they unfold over time interdependently. Lastly, musical
notes are often grouped into chords, arpeggios or melodies in
polyphonic music, and thereby introducing a chronological ordering
of notes is not naturally suitable. this project focuses on building
a GAN Model for Music Generation from Lakh Pypiano Roll dataset
using specifically DCGAN to explore the use of DCGAN instead of
conventional images and text dataset problems.
|
|
20 |
光電碩一 |
Recognize Handwritten Numbers
|
Recognize Handwritten Numbers
|
|