自駕船計畫:辨識愛河上的船及障礙物,將辨識資訊即時回傳至操控台進行自動避碰。
工作項目:
(1) 物件辨識:利用yolo模型進行河面上之障礙物辨識
(2) 物件追蹤:利用深度學習追蹤演算法(Siames RPN)進行高速追蹤
(3) 追蹤優化:辨識結果結合Kalmam filter進行優化
youtube字幕資料收集:將內影片嵌字幕轉換成文字,生成語音訓練資料
工作項目:
(1) 資料爬蟲:使用scrapy進行字幕與影像爬蟲
(2) 字幕辨識:利用super resolution模型將影片背景黑化,再利用OCR將內嵌字幕轉換成文字
self driving boat: return the obstacle location on the love river to the monitoring station
work items:
(1) object detection:use yolo to run the real time object detection
(2) object tracking:implement Siames RPN for high speed object tracking
(3) tracking optimization:combine the Kalmam filter to improve the accuracy of the result of object detection
youtube video subtitle data collection: transform the subtitle image to the txt file
work items:
(1) data collection:use scrapy framework for web crawler
(2) subtitle transform:image preprocess by super resolution model and get the txt file with the OCR technology
深度學習工程師 deep-learning-engineer
清華大學 National Tsing Hua University
Oct 2017 ~ Sep 2018
・
1 yr 0 mos
實驗室要導入時間序列相關深度學習之先導研究,題目為自然語言處理,利用tensorflow搭建NLP語言模型
(1) 文本分類:利用RNN與CNN進行文本分類。
(2) Word to vector:文字向量轉化
(3) 文本摘要總結:利用sequence to sequence模型架構,輸入經由encoder壓縮,再經由decoder解碼進行文本摘要總結訓練。
(4) 對話機器人:收集大量影集對白作為訓練資料,利用sequence to sequence模型架構進行對話機器人訓練。
Study about the time series deep learning model, build RNN model with tensorflow for some natural language processing task
(1) texture classification:texture classification by CNN and RNN network
(2) word to vector:trans the word to the vector domain
(3) text summarization :train text summarization model with encoder-decoder architecture
(4) chat bot:train RNN model for the chat bot with the sequence to sequence architecture
Android app developer
wemo scooter
Oct 2016 ~ Aug 2017
・
11 mos
公司營運項目為電動機車租借,主要開發項目為:
(1) 依照需求進行Android App開發與修改
(2) 與後端進行restful api串接
(3) 串接google map api
(1) Maintain and improve current products
(2) experience with restful api
(3) experience with google map api