logo Idiap Research Institute        
 [BibTeX] [Marc21]
Processing Megapixel Images with Deep Attention-Sampling Models
Type of publication: Idiap-RR
Citation: Katharopoulos_Idiap-RR-07-2019
Number: Idiap-RR-07-2019
Year: 2019
Month: 7
Institution: Idiap
Abstract: Existing deep architectures cannot operate on very large signals such as megapixel images due to computational and memory constraints. To tackle this limitation, we propose a fully differentiable end-to-end trainable model that samples and processes only a fraction of the full resolution input image. The locations to process are sampled from an attention distribution computed from a low resolution view of the input. We refer to our method as attention sampling and it can process images of several megapixels with a standard single GPU setup. We show that sampling from the attention distribution results in an unbiased estimator of the full model with minimal variance, and we derive an unbiased estimator of the gradient that we use to train our model end-to-end with a normal SGD procedure. This new method is evaluated on three classification tasks, where we show that it allows to reduce computation and memory footprint by an order of magnitude for the same accuracy as classical architectures. We also show the consistency of the sampling that indeed focuses on informative parts of the input images.
URL: http://proceedings.mlr.press/v...
Projects Idiap
Authors Katharopoulos, Angelos
Fleuret, Francois
Crossref by Katharopoulos_ICML_2019
Added by: [ADM]
Total mark: 0
  • Katharopoulos_Idiap-RR-07-2019.pdf (MD5: 8a969dedf76a0aa423935bca5949c665)
       (ICML 2019 camera ready with supplementary)