Added images and styles to articles

AscendNTNU · May 23, 2017 · e2f3ce6 · e2f3ce6
1 parent 0ed1f96
commit e2f3ce6
Show file tree

Hide file tree

Showing 4 changed files with 12 additions and 7 deletions.
diff --git a/posts/2016-02-08-Tech-How-We-See-1.md b/posts/2016-02-08-Tech-How-We-See-1.md
@@ -1,7 +1,8 @@
 ---
 layout: post
-title:  "Tech: How we see the world (Part 1)"
-date:   2016-02-08 10:00:00
+image: /public/assets/tech-how-we-see-1/tileframe-4.png
+title: "Tech: How we see the world (Part 1)"
+date: 2016-02-08 10:00:00
 categories: tech computer vision
 author: Simen Haugo
 ---
@@ -72,7 +73,9 @@ Unfortunately, I skipped over a bunch of technical details in the above explanat
 Here's a video I recorded at our lab the other day. In an attempt to somewhat reproduce the competition at home, we have not only bought in a handful of vacuum cleaner robots, but we have also created an elaborate grid of sticky tape.
 
 <figure>
-  <video src="/public/assets/tech-how-we-see-1/scaled.mp4" width="480" height="270" controls></video>
+  <video width="480" height="270" controls>
+    <source src="/public/assets/tech-how-we-see-1/scaled.mp4" type="video/mp4" />
+  </video>
 </figure>
 
 This is, part, what our drone will be seeing. We think we need more cameras, but we're pretty sure we'll have one pointing downwards like this. Naturally, you have no problem working out how big the grid was, or how many tiles I walked in that video. But that's thanks to the many years of intense grid-position-estimation training you've done, right?

diff --git a/posts/2016-08-28-Retrospect-2016.md b/posts/2016-08-28-Retrospect-2016.md
@@ -1,7 +1,8 @@
 ---
 layout: post
-title:  "IARC Postmortem: 2016 in retrospect"
-date:   2016-08-28 12:00:00
+image: /public/assets/retrospect-16/atlanta.jpg
+title: "IARC Postmortem: 2016 in retrospect"
+date: 2016-08-28 12:00:00
 categories: IARC competition
 author: Simen Haugo
 ---

diff --git a/posts/2017-03-10-Perception-Update.md b/posts/2017-03-10-Perception-Update.md
@@ -15,7 +15,7 @@ The figure shows the most basic artificial neuron with the activation function f
 Those artificial neurons are in fact, when put together in unimaginable amounts in very complex networks, the core components that can learn visual features in pictures. When inputting an RGB image in the network for which we know what the network should output, what basically happens is that the weight in each neurons is appropriately adjusted, so that they can learn to fire on pixel values such that the networks output comes close to the desired output. The desired output is often confidence values between [0,1] for each of the classes.
 
 <figure>
-  <img alt"Network Graph" src="/public/assets/perception-update-1/network-graph.jpg" />
+  <img alt="Network Graph" src="/public/assets/perception-update-1/network-graph.jpg" />
   <figcaption>A decription of a fully connected(hidden) layer.</figcaption>
 </figure>
 
@@ -40,7 +40,7 @@ Out of the pit then came the Single Shot Multibox Detection Network, which is ba
 Following the nature of a CNN, when doing convolutions on the features, the network will learn features on increasingly larger scale (in the input picture), the more deeper layer you are into the network. In the picture, the dog is detected later in the network than the cat. We have previously succeeded running a 200 classes distinguishing caffe version of the SSD network on a NVIDIA Jetson TX1 card on 6 fps. The modified SSD network will hopefully run with higher fps, as we only need 2 classes (red, green gr. r.).
 
 <figure>
-  <img alt"SSD VS. YOLO" src="/public/assets/perception-update-1/ssd-vs-yolo.jpg" />
+  <img alt="SSD VS. YOLO" src="/public/assets/perception-update-1/ssd-vs-yolo.jpg" />
   <figcaption>You Only Look Once (YOLO) network, a similar much appraised one, is shown for comparison.</figcaption>
 </figure>
 

diff --git a/server.js b/server.js
@@ -622,6 +622,7 @@ function createFBInstantArticle (data) {
     <link rel="canonical" href="https://ascendntnu.no/blog/${data.link}" />
     <link rel="stylesheet" href="https://fonts.googleapis.com/css?family=Open+Sans|Questrial|Roboto+Slab" />
     <meta property="op:markup_version" content="v1.0">
+    <meta property="fb:article_style" content="Ascend Dark Theme">
     <script type="application/ld+json">
       {
         "@context": "http://schema.org",