(PAGE UNDER DEVELOPMENT)
COFW is "is designed to benchmark face landmark algorithms in realistic conditions, which include heavy occlusions and large shape variations" [Robust face landmark estimation under occlusion].
RESEARCH below this line
We asked four people with different levels of computer vision knowledge to each collect 250 faces representative of typical real-world images, with the clear goal of challenging computer vision methods. The result is 1,007 images of faces obtained from a variety of sources.
Robust face landmark estimation under occlusion
Our face dataset is designed to present faces in real-world conditions. Faces show large variations in shape and occlusions due to differences in pose, expression, use of accessories such as sunglasses and hats and interactions with objects (e.g. food, hands, microphones, etc.). All images were hand annotated in our lab using the same 29 landmarks as in LFPW. We annotated both the landmark positions as well as their occluded/unoccluded state. The faces are occluded to different degrees, with large variations in the type of occlusions encountered. COFW has an average occlusion of over 23%. To increase the number of training images, and since COFW has the exact same landmarks as LFPW, for training we use the original non-augmented 845 LFPW faces + 500 COFW faces (1345 total), and for testing the remaining 507 COFW faces. To make sure all images had occlusion labels, we annotated occlusion on the available 845 LFPW training images, finding an average of only 2% occlusion.
http://www.vision.caltech.edu/xpburgos/ICCV13/
This research is supported by NSF Grant 0954083 and by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), via IARPA R&D Contract No. 2014-14071600012.
To understand how COFW Dataset has been used around the world... affected global research on computer vision, surveillance, defense, and consumer technology, the and where this dataset has been used the locations of each organization that used or referenced the datast
The data is generated by collecting all citations for all original research papers associated with the dataset. Then the PDFs are then converted to text and the organization names are extracted and geocoded. Because of the automated approach to extracting data, actual use of the dataset can not yet be confirmed. This visualization is provided to help locate and confirm usage and will be updated as data noise is reduced.
Citations were collected from Semantic Scholar, a website which aggregates and indexes research papers. Metadata was extracted from these papers, including extracting names of institutions automatically from PDFs, and then the addresses were geocoded. Data is not yet manually verified, and reflects anytime the paper was cited. Some papers may only mention the dataset in passing, while others use it as part of their research methodology.
Add button/link to download CSV
This bar chart presents a ranking of the top countries where citations originated. Mouse over individual columns to see yearly totals. Colors are only assigned to the top 10 overall countries.
TODO