We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CV

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: Bridging the gap between Natural and Medical Images through Deep Colorization

Abstract: Deep learning has thrived by training on large-scale datasets. However, in many applications, as for medical image diagnosis, getting massive amount of data is still prohibitive due to privacy, lack of acquisition homogeneity and annotation cost. In this scenario, transfer learning from natural image collections is a standard practice that attempts to tackle shape, texture and color discrepancies all at once through pretrained model fine-tuning. In this work, we propose to disentangle those challenges and design a dedicated network module that focuses on color adaptation. We combine learning from scratch of the color module with transfer learning of different classification backbones, obtaining an end-to-end, easy-to-train architecture for diagnostic image recognition on X-ray images. Extensive experiments showed how our approach is particularly efficient in case of data scarcity and provides a new path for further transferring the learned color information across multiple medical datasets.
Subjects: Computer Vision and Pattern Recognition (cs.CV); Medical Physics (physics.med-ph)
Cite as: arXiv:2005.10589 [cs.CV]
  (or arXiv:2005.10589v1 [cs.CV] for this version)

Submission history

From: Lia Morra [view email]
[v1] Thu, 21 May 2020 12:03:14 GMT (5653kb,D)

Link back to: arXiv, form interface, contact.