Features

The following tasks are supported:

• Classification - binary classification of the presence of glasses and their types.

• Detection - binary detection of worn/standalone glasses and eye area.

• Segmentation - binary segmentation of glasses and their parts.

Each task has multiple kinds (task categories) and model sizes (architectures with pre-trained weights).

Classification

Classification Kinds

Kind	Description	Examples
anyglasses	Identifies any kind of glasses, googles, or spectacles.
		Positive	Positive	Negative
eyeglasses	Identifies only transparent glasses (here referred as eyeglasses)
		Positive	Negative	Negative
sunglasses	Identifies only opaque and semi-transparent glasses (here referred as sunglasses)
		Negative	Positive	Negative
shadows	Identifies cast shadows (only shadows of (any) glasses frames)
		Positive	Negative	Negative

Check classifier performances

• Performance Information of the Pre-trained Classifiers: performance of each kind.

• Size Information of the Pre-trained Classifiers: efficiency of each size.

Detection

Detection Kinds

Kind	Description	Examples
eyes	Detects only the eye region, no glasses.
solo	Detects any glasses in the wild, i.e., standalone glasses that are placed somewhere.
worn	Detects any glasses worn by people but can also detect non-worn glasses.

Check detector performances

• Performance Information of the Pre-trained Detectors: performance of each kind.

• Size Information of the Pre-trained Detectors: efficiency of each size.

Segmentation

Segmentation Kinds

Kind	Description	Examples
frames	Segments frames (including legs) of any glasses
full	Segments full glasses, i.e., lenses and the whole frame
legs	Segments only frame legs of standalone glasses
lenses	Segments lenses of any glasses (both transparent and opaque).
shadows	Segments cast shadows on the skin by the glasses frames only (does not consider opaque lenses).
smart	Segments visible glasses parts: like full but does not segment lenses if they are transparent.

Check segmenter performances

• Performance Information of the Pre-trained Segmenters: performance of each kind.

• Size Information of the Pre-trained Segmenters: efficiency of each size.