Interactive Learning for Semantic Segmentation in Earth Observation
Overview & Synopsis
Dense pixel-wise classification maps output by deep neuralnetworks are of extreme importance for scene understanding. However,these maps are often partially inaccurate due to a variety of possible fac-tors. Therefore, we propose to interactively refine them within a frame-work named DISCA (Deep Image Segmentation with Continual Adap-tation). It consists of continually adapting a neural network to a targetimage using an interactive learning process with sparse user annotationsas ground-truth. We show through experiments on three datasets usingsynthesized annotations the benefits of the approach, reaching an IoUimprovement up to 4.7% for ten sampled clicks. Finally, we exhibit thatour approach can be particularly rewarding when it is faced to additionalissues such as domain adaptation.
hal.science)
← / → for pages • + / - for zoom • 0 to fit width
Powered by Mozilla PDF.js v3.11
Document Information
| Document Title | Interactive Learning for Semantic Segmentation in Earth Observation |
|---|---|
| Author / Organization | Gaston Lenczner, Adrien Chan-Hon-Tong, Nicolas Luminari, Bertrand Le Saux, Guy Le Besnerais |
| Subject & Discipline | Multidisciplinary Academic Open Research |
| Categories | Science & Mathematics |
| Publication Year | 2020 |
| License / Rights | Open Access / Creative Commons |
| Resource Verification | Listed PDF (Verification Pending) |
| Original Host Website |
HAL Open Science
(hal.science)
|
| File Format | application/pdf (Portable Document Format) |
| Language | EN |
| Catalog ID | DOC-170699 |
| Original Source Link | Visit Source Page on hal.science » |