brainwash, example sidebar

6 files changed, 31 insertions, 10 deletions

diff --git a/site/content/pages/datasets/brainwash/assets/00818000_640x480.jpg b/site/content/pages/datasets/brainwash/assets/00818000_640x480.jpg
deleted file mode 100644
index 30c0fcb1..00000000
--- a/site/content/pages/datasets/brainwash/assets/00818000_640x480.jpg
+++ /dev/null
diff --git a/site/content/pages/datasets/brainwash/assets/background_540.jpg b/site/content/pages/datasets/brainwash/assets/background_540.jpg
deleted file mode 100644
index 5c8c0ad4..00000000
--- a/site/content/pages/datasets/brainwash/assets/background_540.jpg
+++ /dev/null
diff --git a/site/content/pages/datasets/brainwash/assets/background_600.jpg b/site/content/pages/datasets/brainwash/assets/background_600.jpg
deleted file mode 100755
index 8f2de697..00000000
--- a/site/content/pages/datasets/brainwash/assets/background_600.jpg
+++ /dev/null
diff --git a/site/content/pages/datasets/brainwash/assets/brainwash_mean_overlay.jpg b/site/content/pages/datasets/brainwash/assets/brainwash_mean_overlay.jpg
new file mode 100755
index 00000000..2f5917e3
--- /dev/null
+++ b/site/content/pages/datasets/brainwash/assets/brainwash_mean_overlay.jpg
diff --git a/site/content/pages/datasets/brainwash/assets/brainwash_mean_overlay_wm.jpg b/site/content/pages/datasets/brainwash/assets/brainwash_mean_overlay_wm.jpg
new file mode 100755
index 00000000..790dbb79
--- /dev/null
+++ b/site/content/pages/datasets/brainwash/assets/brainwash_mean_overlay_wm.jpg
diff --git a/site/content/pages/datasets/brainwash/index.md b/site/content/pages/datasets/brainwash/index.md
index 0bf67455..d9bffb39 100644
--- a/site/content/pages/datasets/brainwash/index.md
+++ b/site/content/pages/datasets/brainwash/index.md
@@ -19,28 +19,24 @@ authors: Adam Harvey
 + Published: 2015
 + Images: 11,918
 + Faces: 91,146
-+ Created by: Stanford Department of Computer Science
++ Created by: Stanford University (US)<br>Max Planck Institute for Informatics (DE)
 + Funded by: Max Planck Center for Visual Computing and Communication
-+ Location: Brainwash Cafe, San Franscisco
-+ Purpose: Training face detection
++ Purpose: Face detection
 + Website: <a href="https://exhibits.stanford.edu/data/catalog/sx925dc9385">stanford.edu</a>
-+ Paper: <a href="http://arxiv.org/abs/1506.04878">End-to-End People Detection in Crowded Scenes</a>
-+ Explicit Consent: No
 
 
 ## Brainwash Dataset
 
 (PAGE UNDER DEVELOPMENT)
 
-*Brainwash* is a face detection dataset created from the Brainwash Cafe's livecam footage including 11,918 images of "everyday life of a busy downtown cafe[^readme]". The images are used to develop face detection algorithms for the "challenging task of detecting people in crowded scenes" and tracking them.
+*Brainwash* is a head detection dataset created from San Francisco's Brainwash Cafe livecam footage. It includes 11,918 images of "everyday life of a busy downtown cafe[^readme]". The images are used to train and validate algorithms for detecting people in crowded scenes.
 
-Before closing in 2017, Brainwash Cafe was a "cafe and laundromat" located in San Francisco's SoMA district. The cafe published a publicy available livestream from the cafe with a view of the cash register, performance stage, and seating area.
+Before closing in 2017, The Brainwash Cafe was a combination cafe, laundromat, and performance venue located in San Francisco's SoMA district. The images used for Brainwash dataset were captured on 3 days: October 27, November 13, and November 24 in 2014. According the author's reserach paper introducing the dataset, the images were acquired with the help of Angelcam.com [cite orig paper].
 
-Since it's publication by Stanford in 2015, the Brainwash dataset has appeared in several notable research papers. In September 2016 four researchers from the National University of Defense Technology in Changsha, China used the Brainwash dataset for a research study on "people head detection in crowded scenes", concluding that their algorithm "achieves superior head detection performance on the crowded scenes dataset[^localized_region_context]". And again in 2017 three researchers at the National University of Defense Technology used Brainwash for a study on object detection noting "the data set used in our experiment is shown in Table 1, which includes one scene of the brainwash dataset[^replacement_algorithm]".
+Brainwash is not a widely used dataset but since it's publication by Stanford in 2015, it has notably appeared in several research papers from the National University of Defense Technology in Changsha, China. In 2016 and in 2017 researchers there conducted studies on "people head detection in crowded scenes" [^localized_region_context] [^replacement_algorithm].
 
-![caption: An sample image from the Brainwash dataset used for training face and head detection algorithms for surveillance. The datset contains about 12,000 images. License: Open Data Commons Public Domain Dedication (PDDL)](assets/00425000_960.jpg)
+![caption: The pixel-averaged image of all Brainwash dataset images is shown with 81,973 head annotations drawn from the Brainwash training partition. (c) Adam Harvey](assets/brainwash_mean_overlay.jpg)
 
-![caption: 49 of the 11,918 images included in the Brainwash dataset. License: Open Data Commons Public Domain Dedication (PDDL)](assets/brainwash_montage.jpg)
 
 {% include 'chart.html' %}
 
@@ -55,12 +51,37 @@ Add more analysis here
 
 {% include 'citations.html' %}
 
+![caption: An sample image from the Brainwash dataset used for training face and head detection algorithms for surveillance. The datset contains about 12,000 images. License: Open Data Commons Public Domain Dedication (PDDL)](assets/00425000_960.jpg)
+
+![caption: 49 of the 11,918 images included in the Brainwash dataset. License: Open Data Commons Public Domain Dedication (PDDL)](assets/brainwash_montage.jpg)
 
 ### Additional Information
 
 - The dataset author spoke about his research at the CVPR conference in 2016 <https://www.youtube.com/watch?v=Nl2fBKxwusQ>
 
 
+To evaluate the performance of our approach, we collected a
+large dataset of images from busy scenes using video footage available from public webcams. In
+total, we collect 11917 images with 91146 labeled people. We extract images from video footage at
+a fixed interval of 100 seconds to ensure a large variation in images. We allocate 1000 images for
+testing and validation, and leave the remaining images for training, making sure that no temporal
+overlaps exist between training and test splits. The resulting training set contains 82906 instances.
+Test and validation sets contain 4922 and 3318 people instances respectively. Images were labeled
+using Amazon Mechanical Turk by a handful of workers pre-selected through their performance on
+an example task. We label each person’s head to avoid ambiguity in bounding box locations. The
+annotator labels any person she is able to recognize, even if a substantial part of the person is not
+visible. Images and annotations will be made available 1 .
+Examples of collected images are shown in Fig. 6, and in the video included in the supplemental
+material. Images in our dataset include challenges such as people at small scales, strong partial
+occlusions, and a large variability in clothing and appearance.
+
+TODO
+
+- add bounding boxes to the header image
+- remake montage with randomized images, with bboxes
+- clean up intro text
+
+
 ### Footnotes
 
 [^readme]: "readme.txt" https://exhibits.stanford.edu/data/catalog/sx925dc9385.