EURASIP Journal on Image and Video Processing

Table 1 Selection of publicly available facial landmarking databases

Name	Images	Subjects	Landmarks	Description	Year
XM2VTS [77]	2 360	295	68pts^a	Controlled, 2 head	1999
	(720 × 560)			rotation images, 6 images
	RGB			captured during speech
BioID [82]	1 521	23	20	Controlled, front on image,	2001
	(384 × 286)			varying expressions
	Grayscale
LFW [84]	13 233	5749 (1680	MULTI-PIE	Uncontrolled, originally	2007
	(250 × 250)	have ≥ 2 images	68pts^a	intended for face identification,
	mostly RGB	in the set)		images collected from the web
Caltech [81]	7 092 varied size	Unknown (10	4pts	Uncontrolled, images collected	2007
10 000	mostly RGB	524 faces in		from Google images
web faces		7 092 images)
PUT [80]	9 971	100	30 primary	Controlled,	2008
	(2048 × 1536) RGB		(194 control	portrait images,
			on subset)	5 poses per subject
MULTI-PIE [75]	755 370	337	68pts	Controlled,	2008
	(3072 × 2048) RGB			landmarks for subset,
				varying expressions
MUCT [76]	3 755	276	XM2VTS 68pts +4	Controlled, 5 perspectives	2010
	(640 × 480) RGB		around eyes (76 total)	3 lighting conditions
				Neutral expression or smile
AFLW [79]	2 330 varied	Unknown	PUT 97pts^a	Uncontrolled,	2012
	size RGB			portrait images, collected
				from the web (Flickr)
HELEN [83]	2 330	Unknown	PUT 97pts^a	Uncontrolled, collected	2012
	RGB			from the web (Flickr)
300W [10, 11]	600 varied	Unknown	MULTI-PIE 68pts	Uncontrolled, 300 outdoor	2013
	size RGB			images 300 indoor images
				difficult poses and expressions
Menpo benchmark [78]	8 979 varied	Unknown	Frontal: MULTI-PIE	Uncontrolled, 6679 frontal	2017
	size RGB		68 pts Profile: 39pts	images 2300 profile images
				difficult poses and expressions

^aThe Intelligent Behaviour Understanding Group (iBUG) [71] have made MULTI-PIE 68pts landmarks available for this dataset