]({{ site.baseurl }}/)
diff --git a/README.md b/README.md
index ca5d072..45c2020 100644
--- a/README.md
+++ b/README.md
@@ -1,15 +1,120 @@
-# updated website
+> March, 2016: If you're on an old version of Jekyll Now and run into a) build warnings or b) syntax highlighting issues caused by [Jekyll 3 and GitHub Pages updates](https://github.com/blog/2100-github-pages-now-faster-and-simpler-with-jekyll-3-0), just :sparkles:[update your _config.yml](https://github.com/barryclark/jekyll-now/pull/445/files):sparkles: and you'll be set!
-This repo is built on a fork of **Jekyll Now** from [this repository](https://github.com/barryclark/jekyll-now). **Jekyll** is a static site generator that's perfect for GitHub hosted blogs ([Jekyll Repository](https://github.com/jekyll/jekyll))
+# Jekyll Now
-The website design is just a modification of [Jon Barron's website](https://jonbarron.info/) and is converted for my own use, re-purposing my old markdown posts. **Feel free to use template for your own purposes**, but please respect copyright for all the images/content in my `images`, `pdfs`, `_posts` folders.
+**Jekyll** is a static site generator that's perfect for GitHub hosted blogs ([Jekyll Repository](https://github.com/jekyll/jekyll))
+**Jekyll Now** makes it easier to create your Jekyll blog, by eliminating a lot of the up front setup.
+- You don't need to touch the command line
+- You don't need to install/configure ruby, rvm/rbenv, ruby gems :relaxed:
+- You don't need to install runtime dependencies like markdown processors, Pygments, etc
+- If you're on Windows, this will make setting up Jekyll a lot easier
+- It's easy to try out, you can just delete your forked repository if you don't like it
-## issues
-* In general, jekyll will try to build a full page for every post. I skip that by forcing `permalink: /`. This creates multiple entries in sitemap.xml for index.html but is otherwise fine.
-* If you want multiple paragraphs, consider using `excerpt_separator: ` in `_config.yml`, for my own use I didn't need this.
-* My own posts have lots of extra stuff left over from my old jekyll design ("author", long descriptions, etc.), feel free to ignore them
-* I use thumbnails, so I can upload arbitrary sized images but then only display small ones. The `_make_thumbnails.sh` script generates them and the html template looks in `tn/` for all images.
-* I have three categories of post with slightly differerent formatting, so changing sizing requires edits in multiple paces.
-* If you use this, I'd appreciate a link back either to this repo or my personal website so others can find this too.
+In a few minutes you'll be set up with a minimal, responsive blog like the one below giving you more time to spend on writing epic blog posts!
+
+
+
+## Quick Start
+
+### Step 1) Fork Jekyll Now to your User Repository
+
+Fork this repo, then rename the repository to yourgithubusername.github.io.
+
+Your Jekyll blog will often be viewable immediately at 
- {{ page.title }}
+ +
+ {{ content }}
+
+
+
+
+
+
+{% endif %}
\ No newline at end of file
diff --git a/_includes/meta.html b/_includes/meta.html
new file mode 100644
index 0000000..b8f4a33
--- /dev/null
+++ b/_includes/meta.html
@@ -0,0 +1,18 @@
+
+
+
+
+
+ {% if page.excerpt %}
+
+
+ {% else %}
+
+
+ {% endif %}
+
+
+ {% if page.title %}
+
+
+ {% endif %}
\ No newline at end of file
diff --git a/_includes/svg-icons.html b/_includes/svg-icons.html
new file mode 100644
index 0000000..7a64482
--- /dev/null
+++ b/_includes/svg-icons.html
@@ -0,0 +1,13 @@
+{% if site.footer-links.dribbble %}{% endif %}
+{% if site.footer-links.email %}{% endif %}
+{% if site.footer-links.facebook %}{% endif %}
+{% if site.footer-links.flickr %}{% endif %}
+{% if site.footer-links.github %}{% endif %}
+{% if site.footer-links.instagram %}{% endif %}
+{% if site.footer-links.linkedin %}{% endif %}
+{% if site.footer-links.pinterest %}{% endif %}
+{% if site.footer-links.rss %}{% endif %}
+{% if site.footer-links.twitter %}{% endif %}
+{% if site.footer-links.stackoverflow %}{% endif %}
+{% if site.footer-links.youtube %}{% endif %}
+{% if site.footer-links.googleplus %}{% endif %}
\ No newline at end of file
diff --git a/_layouts/default.html b/_layouts/default.html
index 743d933..fd48a90 100644
--- a/_layouts/default.html
+++ b/_layouts/default.html
@@ -1,252 +1,37 @@
-
-
-
-
-
-
- - -
- - -
|
-
+ {{ content }}
+
+
+
+ {% include analytics.html %}
+
+
diff --git a/_layouts/page.html b/_layouts/page.html
new file mode 100644
index 0000000..c8f0164
--- /dev/null
+++ b/_layouts/page.html
@@ -0,0 +1,12 @@
+---
+layout: default
+---
+
+{{ page.title }}
+ +
+ {{ content }}
+
+
+
+ Written on {{ page.date | date: "%B %e, %Y" }}
+
+
+ {% include disqus.html %}
+-diff --git a/_posts/2013-06-07-dashpoint.markdown b/_posts/2013-06-07-dashpoint.markdown deleted file mode 100644 index dfa818d..0000000 --- a/_posts/2013-06-07-dashpoint.markdown +++ /dev/null @@ -1,10 +0,0 @@ ---- -layout: post -title: "DashPoint: A low-cost, low-power human interface device" -date: 2013-06-07 22:21:59 +00:00 -image: /images/dashpoint.png -categories: Intel -patent: https://patents.google.com/patent/US10007994B2 -patent2: https://patents.google.com/patent/US9494415B2 ---- -Finger tracking on a microcontroller, with optics tricks and some HCI ideas \ No newline at end of file diff --git a/_posts/2015-04-08-box.markdown b/_posts/2015-04-08-box.markdown deleted file mode 100644 index 75dd1a4..0000000 --- a/_posts/2015-04-08-box.markdown +++ /dev/null @@ -1,22 +0,0 @@ ---- -layout: post -title: "Real-time Box Measurement" -date: 2015-04-08 21:21:53 +00:00 -image: /images/box.jpg -categories: Intel -author: "Leo Keselman" -subtitle: "Real-time measurement of rectangular prisms" -video: https://www.youtube.com/watch?v=ZZvZVaBFpYE -video2: https://www.youtube.com/watch?v=rYnFWkF7Jx8 ---- -Using a single depth sensor, real-time detection of cuboids, accurate estimation of their dimensions, and even some bin-packing. - -To show that the Intel RealSense cameras were capable of real-time volume measurement. I designed and built an algorithm that decomposed the problem into two parts, a surface-estimation step and a bounding rectangle step. Then, with basic geometry, was able to build a system that estimated object volume, even with only two visible sides. This was further expanded with a configurable-objective [Bin Packing solver](https://en.wikipedia.org/wiki/Bin_packing_problem){:target="_blank"} that I wrote. The use of this is that most shipping companies (FedEx, USPS, UPS, DHL, etc.) now are limited (and hence charge), by volume not weight. - - -With the help of [Sterling Orsten](https://github.com/sgorsten){:target="_blank"} and [Dimitri Diakopoulos](https://github.com/ddiakopoulos){:target="_blank"}, we put together a compelling visual prototype that was shown by Intel's CEO at IDF 2015 (China). - - -[YouTube video of algorithm](https://www.youtube.com/watch?v=rYnFWkF7Jx8){:target="_blank"} - -[YouTube video of IDF 2015 Demo](https://www.youtube.com/watch?v=ZZvZVaBFpYE){:target="_blank"} diff --git a/_posts/2015-04-22-dequant.markdown b/_posts/2015-04-22-dequant.markdown deleted file mode 100644 index 4a0b54e..0000000 --- a/_posts/2015-04-22-dequant.markdown +++ /dev/null @@ -1,14 +0,0 @@ ---- -layout: post -title: "Dequantization of Depth Data" -date: 2015-04-22 22:21:59 +00:00 -image: /images/dequant.jpg -categories: Other -author: "Leo Keselman" -subtitle: "For higher quality normal maps" -code: https://github.com/leonidk/qInterp ---- -An O(1) time algorithm for producing smooth normals for quantized data, such as the Kinect. - - -[GitHub Link](https://github.com/leonidk/qInterp){:target="_blank"} diff --git a/_posts/2015-06-12-cvpr15.markdown b/_posts/2015-06-12-cvpr15.markdown deleted file mode 100644 index e0fb9f3..0000000 --- a/_posts/2015-06-12-cvpr15.markdown +++ /dev/null @@ -1,19 +0,0 @@ ---- -layout: post -title: "Rigid-body Dynamics for Articulated Mesh Tracking" -date: 2015-06-12 22:20:59 +00:00 -image: /images/hands2015.png -categories: research -author: "Leo Keselman" -subtitle: "Invited Workshop Talk" -venue: "CVPR Workshops (HANDS)" -slides: /pdfs/hands2015.pdf -authors: "Leonid Keselman, Sterling Orsten, Stan Melax" ---- -An invited talk for the [HANDS 2015](http://www.ics.uci.edu/~jsupanci/HANDS-2015/){:target="_blank"} workshop at CVPR 2015. This includes further details about the efficiency of our rigid-body solver, our machine-learning tools, and some details about our data annotation process. - -[Presented Slides](/pdfs/hands2015.pdf){:target="_blank"} - -[Video of Fast Tracking](https://www.youtube.com/watch?v=sTOF0eY9uv4){:target="_blank"} - -[Video of Tracking Under Uncertainty](https://www.youtube.com/watch?v=_DogsLiC4XY){:target="_blank"} \ No newline at end of file diff --git a/_posts/2015-12-04-cs279.markdown b/_posts/2015-12-04-cs279.markdown deleted file mode 100644 index 2d273b0..0000000 --- a/_posts/2015-12-04-cs279.markdown +++ /dev/null @@ -1,21 +0,0 @@ ---- -layout: post -title: "Level-set based tracking and segmentation" -date: 2015-12-04 22:22:59 +00:00 -image: /images/cs279.png -categories: Stanford -author: "Leo Keselman" -subtitle: "Level-set based tracking and segmentation" -code: https://github.com/leonidk/drlse -paper: /pdfs/cs279.pdf -course: "CS279: Structure and Organization of Biomolecules and Cells" ---- - -We implemented a detection and deformable tracking pipeline for red blood cells. - - including automated detection with a Hough Circle Transform, and deformable tracking with our own Python implementation of Distance Regularized Level Set Evolution on top of basic image processing primitives provided by Python’s scikit-image library (e.g. Gaussian Filters). We analyze the behavior of this tracking system and its shortcomings. - - -[Paper PDF](/pdfs/cs279.pdf){:target="_blank"} - -[GitHub](https://github.com/leonidk/drlse){:target="_blank"} \ No newline at end of file diff --git a/_posts/2015-12-12-cs221.markdown b/_posts/2015-12-12-cs221.markdown deleted file mode 100644 index f366da3..0000000 --- a/_posts/2015-12-12-cs221.markdown +++ /dev/null @@ -1,30 +0,0 @@ ---- -layout: post -title: "Doctor Bayes" -date: 2015-12-12 22:21:59 +00:00 -image: /images/drbayes.png -categories: Stanford -author: "Leo Keselman" -subtitle: "Symptom-based disease prediction" -course: "CS221: Artificial Intelligence" -code: https://github.com/leonidk/drbayes -website: http://doctorbayes.com -poster: /pdfs/cs229-poster.pdf -paper: /pdfs/cs221.pdf ---- - -Detecting disease from a short description of symptoms. In some small testing, obtained nearly 90% top 5 accuracy and about 60% top 1 accuracy - -For Stanford's Machine Learning class, I worked on a project to alleviate the impact of healthcare spending on the the “last mile” of medicine and predict a user’s illness simply based on a description of their symptoms. While handling multiple online data sources we found that a large part of the challenge was consolidating information across data sources in a consistent matter. We explore several techniques for automatic document matching across data bases of documents,with the goal of finding matching documents across Freebase, Mayo Clinic, andWikipedia. We define matching documents to be articles which describe the same disease, then we found matching documents between databases with 91% accuracy. - -Using data from Freebase, Mayo Clinic, and Wikipedia, we trained a Naive Bayes, Logistic Regression, Random Trees, and many other ML models. We obtain nearly 90% top 5 accuracy and about 60% top 1 accuracy - - -Now hosted at [http://doctorbayes.com](http://doctorbayes.com){:target="_blank"} I rebuilt this project from scratch (without the use of other libraries) in both Python and Javascript. The source code is available on my [github](https://github.com/leonidk/drbayes){:target="_blank"}. - - -[CS221 PDF](/pdfs/cs221.pdf){:target="_blank"} - -[CS229 PDF](/pdfs/cs229.pdf){:target="_blank"} - -[CS229 Poster](/pdfs/cs229-poster.pdf){:target="_blank"} diff --git a/_posts/2016-03-07-cs448i.markdown b/_posts/2016-03-07-cs448i.markdown deleted file mode 100644 index 461c477..0000000 --- a/_posts/2016-03-07-cs448i.markdown +++ /dev/null @@ -1,20 +0,0 @@ ---- -layout: post -title: "Wide-angle Stereo Lenses" -date: 2016-03-07 22:20:59 +00:00 -image: /images/cs448i.png -categories: Stanford -author: "Leo Keselman" -course: "CS448I: Computational Imaging and Display" -subtitle: "Effects of optics on stereo depth generation" -paper: /pdfs/cs448i.pdf -poster: /pdfs/cs448i-poster.pdf ---- - -We introduce various projection functions in the analysis of stereoscopic depth sensors. - -Through these methods, we are able to design stereo systems which are not bound by traditional quadratic depth error. Additionally, we demonstrate how existing correspondence algorithms can be modified to handle these lens designs. In addition, we can construct lens projection models which are more suited to natural lens designs, and build stereo systems with minimal re-sampling errors. This also allows us to construct wide-angler stereoscopic systems than previously possible, without significant re-sampling or sacrificing accuracy. - -[CS448I Paper](/pdfs/cs448i.pdf){:target="_blank"} - -[CS448I Poster](/pdfs/cs448i-poster.pdf){:target="_blank"} \ No newline at end of file diff --git a/_posts/2016-03-08-cs231n.markdown b/_posts/2016-03-08-cs231n.markdown deleted file mode 100644 index d23081b..0000000 --- a/_posts/2016-03-08-cs231n.markdown +++ /dev/null @@ -1,19 +0,0 @@ ---- -layout: post -title: "CNNs for 3D Model Classification" -date: 2016-03-08 22:20:59 +00:00 -image: /images/cs231n.png -categories: Stanford -author: "Leo Keselman" -subtitle: "Learning Projections" -course: "CS231n: Convolutional Neural Networks for Visual Recognition" -poster: /pdfs/cs231n-poster.pdf -paper: /pdfs/cs231n.pdf ---- - -3D shape classification by learning an embedding function into a 2D image and using a pre-trained ImageNet network. At the time, got state-of-the-art results for single-view classification on ShapeNet40. - - -[CS231N Paper](/pdfs/cs231n.pdf){:target="_blank"} - -[CS231N Poster](/pdfs/cs231n-poster.pdf){:target="_blank"} \ No newline at end of file diff --git a/_posts/2016-06-06-cs224u.markdown b/_posts/2016-06-06-cs224u.markdown deleted file mode 100644 index d94a903..0000000 --- a/_posts/2016-06-06-cs224u.markdown +++ /dev/null @@ -1,20 +0,0 @@ ---- -layout: post -title: "Multimodal Natural Language Inference" -date: 2016-06-06 22:20:59 +00:00 -image: /images/cs224u.png -categories: Stanford -author: "Leo Keselman" -subtitle: "CNNs meet RNNs" -paper: /pdfs/cs224u.pdf -course: "CS224U: Natural Language Understanding" -video: https://www.youtube.com/watch?v=WO1RJC_9k7s ---- - -We explored how natural language inference tasks can be augmented with visual data. - -Namely, we replicate and expand existing baselines for NLI, including recent deep learning methods. By adding image features to these models, we explore how the textual and visual modalities interact. Specifically, we show that image features can provide a small boost in classifier performance for simpler models, but are a subset of information provided in the premise statement and thus do not benefit complex models. Additionally, we demonstrate a weakness in the SNLI dataset, showing that textual entailment is predictable without reference to the premise statement. - -[CS224U Paper](/pdfs/cs224u.pdf){:target="_blank"} - -[YouTube Presentation](https://www.youtube.com/watch?v=WO1RJC_9k7s){:target="_blank"} \ No newline at end of file diff --git a/_posts/2016-06-07-cs231a.markdown b/_posts/2016-06-07-cs231a.markdown deleted file mode 100644 index c08b978..0000000 --- a/_posts/2016-06-07-cs231a.markdown +++ /dev/null @@ -1,20 +0,0 @@ ---- -layout: post -title: "Gradient-learned Models for Stereo Matching" -date: 2016-06-07 22:20:59 +00:00 -image: /images/cs231a.png -categories: Stanford -author: "Leo Keselman" -subtitle: "Classification for Stereo Matching" -paper: /pdfs/cs231a.pdf -course: "CS231A: Computer Vision, From 3D Reconstruction to Recognition" -code: https://github.com/leonidk/centest ---- - -Some re-implementations of standard stereo correspondence algorithms, along with experiments using classification for stereo matching. - -In this project, we are exploring the application of machine learning to solving the classical stereoscopic correspondence problem. We present a re-implementation of several state-of-the-art stereo correspondence methods. Additionally, we present new methods, replacing one of the state-of-the-art methods for stereo with a proposed technique based on machine learning methods. These new methods out-perform existing heuristic baselines significantly. - -[CS231A Paper](/pdfs/cs231a.pdf){:target="_blank"} - -[Source Code](https://github.com/leonidk/centest){:target="_blank"} \ No newline at end of file diff --git a/_posts/2016-06-27-vcsel.markdown b/_posts/2016-06-27-vcsel.markdown deleted file mode 100644 index 97ca714..0000000 --- a/_posts/2016-06-27-vcsel.markdown +++ /dev/null @@ -1,11 +0,0 @@ ---- -layout: post -title: "Compact VCSEL Projector" -date: 2016-06-27 22:21:59 +00:00 -image: /images/vcsel.png -categories: Intel -patent: https://patents.google.com/patent/US10007994B2 -patent2: https://patents.google.com/patent/US20170374244A1 -patent3: https://patents.google.com/patent/US9992474B2 ---- -A low-cost dense, configurable projector system for RGB-D depth sensors. \ No newline at end of file diff --git a/_posts/2016-08-06-depth-enhance.markdown b/_posts/2016-08-06-depth-enhance.markdown deleted file mode 100644 index a9b0531..0000000 --- a/_posts/2016-08-06-depth-enhance.markdown +++ /dev/null @@ -1,16 +0,0 @@ ---- -layout: post -title: " Depth Image Enhancement" -date: 2015-08-06 22:21:59 +00:00 -image: /images/haowei.png -categories: Intel -author: "Leo Keselman" -subtitle: "Hand Tracking Demo" -patent: https://patents.google.com/patent/US9661298B2 ---- -Algorithms to filter, enhance and clean-up RGB-D data streams. - - system's architecture, development and the implementation of necessary computer vision algorithms. -- This research explores a new approach to tracking hands, or any articulated model, by using an augmented rigid body simulation. This allows us to phrase 3D object tracking as a linear complementarity problem with a well-defined solution. Based on a depth sensor’s samples, the system generates constraints that limit motion orthogonal to the rigid body model’s surface. These constraints, along with prior motion, collision/contact constraints, and joint mechanics, are resolved with a projected Gauss-Seidel solver. Due to camera noise properties and attachment errors, the numerous surface constraints are impulse capped to avoid overpowering mechanical constraints. To improve tracking accuracy, multiple simulations are spawned at each frame and fed a variety of heuristics, constraints and poses. A 3D error metric selects the best-fit simulation, helping the system handle challenging hand motions. -
-
+
+
+
+
+
+
+
+
+
+
+
+Xiaodong Cun
+Ph.D. candidate at University of Macau
+Department of Computer and Information Science
+Email / Google Scholar / Github
+
+Biography
+
+ I am a Ph.D. student at University of Macau, under the supervision of Dr. Chi-Man Pun.
+ Before that, I received my B.Sc and M.Sc. degrees in Computer Science from Xidian University and University of Macau in 2015 and 2018 respectively.
+ Currently, I am focusing on the application of Deep Learning based methods for Computer Vision and Computational Photograpy.
+
+
+
+
+
+
++
News
+-
+
- One paper is accepted by AAAI-21. 2020.12
+
- One paper is accepted by ECCV 2020. 2020.07
+
- I will join Huawei Noah's Ark Lab as a research intern. 2020.03
+
- One paper is accepted by TIP! 2020.02
+
- One paper is accepted by AAAI-20! 2019.11
+
Activities
+-
+
+
- Intern FrondEnd Engineer, Alibaba Group, Hangzhou, China 2014.7-9
+
- Intern Algorithm Engineer, HIKVision Research Institute, HangZhou, China 2018.5-7
+
- Asian University Machine Learning Camp, Jeju National University, Jeju, Korea 2018.9
+
- Research Intern, Huawei Noah's Ark Lab, Shenzhen, China 2020.3-11
+
Selected Publications
+
+
+ {% for post in site.posts %}
+
+
+ 
+
+
+ {% endfor %}
+
+
+
+
+{{ post.title }}
+
+
+ {{ post.authors }}
+
+
+
+ {{post.publication}}, {{post.year}}
+ {% if post.journal%}
+ /
+ {{post.journal}}, {{post.year}}
+ {% endif %}
+
+
+
+
+
+
+
+
+ {% if post.arxiv%}
+ arXiv
+ {% endif %}
+
+ {% if post.demo%}
+ /
+ Online Demo
+ {% endif %}
+
+ {% if post.code%}
+ /
+ Github
+ {% endif %}
+
+
+
+ Misc
+-
+
- If you are looking for Machine Tranlation papers, please checkout my friend Xuebo Liu. +