D
Diane Larlus
Researcher at Xerox
Publications - 82
Citations - 6174
Diane Larlus is an academic researcher from Xerox. The author has contributed to research in topics: Computer science & Object (computer science). The author has an hindex of 27, co-authored 69 publications receiving 4722 citations. Previous affiliations of Diane Larlus include Technische Universität Darmstadt & Naver Corporation.
Papers
More filters
Proceedings Article
CATEGORY LEVEL OBJECT SEGMENTATION - Learning to Segment Objects with Latent Aspect Models
Diane Larlus,Frédéric Jurie +1 more
TL;DR: A new method for learning to segment objects in images based on a latent variables model used for representing images and objects, inspired by the LDA model, which is extremely well suited for assigning image patches to objects, and therefore for segmenting objects.
Proceedings ArticleDOI
On the use of regions for semantic image segmentation
TL;DR: It is shown that the proposed region based system can achieve good results without any complex regularization, while its patch based counterpart becomes competitive when using image prior and regularization methods.
Category Level Object Segmentation by Combining Bag-of-words Models and Markov Random Fields
TL;DR: The proposed method successfully segments object categories, despite highly varying appearances, cluttered backgrounds and large viewpoint changes, in a probabilistic model based on a bag-of-words representation.
Journal ArticleDOI
Improving the Generalization of Supervised Models
TL;DR: This paper enrichs the common supervised training framework using two key components of recent SSL models: multi-scale crops for data augmentation and the use of an expendable projector head, and replaces the last layer of class weights with class prototypes computed on the fly using a memory bank.
Proceedings Article
No Reason for No Supervision: Improved Generalization in Supervised Models
TL;DR: T-ReX as discussed by the authors proposes a supervised learning setup that leverages the best of both worlds for transfer learning by using multi-scale crops for data augmentation and an expendable projector head to control the tradeoff between performance on the training task and transferability.