Departamento de Informática e
ingeniería de sistemas (DIIS)
Escuela de Ingeniería y
Universidad de Zaragoza (UZ)
C/Maria de Luna, 1
50018 Zaragoza, Spain
Phone: (+34) 897 55 50 75
Email: acambra | AT | unizar |
DOT | es
How to adapt existing CNN models for new semantic segmentation tasks?
need to learn new specific classes but do not have a lot of labeled
training data. This work addresses the problem of transferring the
knowledge from existing CNN models to different classes to those the
model has been trained for. Our work explores the two common transfer
learning approaches for the particular problem of semantic
1) fine-tuning existing models with the newtraining data, following a standard pipeline;
2) training a superpixel classifier combining local and context information.
A generic tool for interactive complex image editing
This work presents
an efficient interaction paradigm that approximates any per-pixel
magnitude from a few user strokes by propagating the sparse user input
to each pixel of the image. We illustrate this through three
interactive applications: depth of field simulation, dehazing and tone
A generic tool for interactive complex image editing. A. B. Cambra, A. C.
Murillo, A. Muñoz. The Visual Computer 2017, p. 1-13. [PDF] [bibtex][Project page]
How to adapt existing CNN models for new semanticsegmentation tasks?A. B. Cambra, A. Muñoz, A. C.
Murillo. Robot 2017: Thrid Iberian Robotics Conference. Sevilla, Spain. [PDF]