We attack the problem of learning face models for public faces from weakly-labelled
images collected from web through querying a name. The data is very noisy even after
face detection, with several irrelevant faces corresponding to other people. We propose
a novel method, Face Association through Model Evolution (FAME), that is able to
prune the data in an iterative way, for the face models associated to a name to evolve. the
idea is based on capturing discriminativeness and representativeness of each instance and
eliminating the outliers. FAME is a generic method that can be used in different domains even we propose it to learn face models from noisy web images. The final models are used to classify faces on novel datasets with possibly different characteristics. On benchmark datasets, our results are comparable to
or better than state-of-the-art studies for the task of face identification.
FAME iterations in noise injected PubFig83 Adam Sandler category
Released Datasets --
For a given name, 500 images are gathered using Bing image search
1 . Categories are chosen as the people having more than 50 annotated face images in FAN-large or PubFig83 datasets. In total, 226691 images are collected corresponding to 365
name categories in FAN-large, and 83 name categories in PubFig83.
2500 faceimages for queries "female face", "male face", "face images" are collected to construct the
Comperative results with the models learned from Web images. For details refer to the paper.
Comperative results of our model curating pipeline. For details refer to the paper.