SALISA: Saliency-based Input Sampling for Efficient Video Object Detection

[ECCV22] We propose SALISA, a novel non-uniform SALiency-based Input SAmpling technique for video object detection that allows for heavy down-sampling of unimportant background regions while preserving the fine-grained details of a high-resolution image. The resulting image is spatially smaller, leading to reduced computational costs while enabling a performance comparable to a high-resolution input. •

FrameExit: Conditional Early Exiting for Efficient Video Recognition

[CVPR21 Oral] In this paper, we present FrameExit, a conditional early exiting method for efficient video recognition. Our proposed method uses gating modules, trained to allow the network to automatically determine the earliest exiting point based on the inferred complexity of the input video. •

Skip-Convolutions for Efficient Video Processing

[CVPR21] We propose Skip-Convolutions to leverage the large amount of redundancies in video streams and save computations. We reformulate standard convolution to be efficiently computed on residual frames: each layer is coupled with a binary gate deciding whether a residual is important to the model prediction, e.g. foreground regions, or it can be safely skipped, e.g. background regions. •

Conditional Channel Gated Networks for Task-Aware Continual Learning

[CVPR20 Oral] We present a framework based on conditional computation to tackle catastrophic forgetting in convolutional neural networks for task-aware continual learning. •

Batch-shaping for learning conditional channel gated networks

[ICLR20 Spotlight] We propose a method that trains large-capacity neural networks with significantly improved accuracy and lower dynamic computational cost. •

Using deep convolutional neural networks to identify and classify tumor-associated stroma in diagnostic breast biopsies

My work on deep learning based analysis of breast tumor stroma in collaboration with Harvard Medical school, NCI, and Mayo Clinic. •

Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer

In this research, we investigated and compared the discriminative accuracy of deep learning algorithms with the diagnoses of pathologists in detecting lymph node metastases in tissue sections of women with breast cancer.

Our work has been published in the Journal of the American Medical Association and is available here.

Our work has more than 2000 citations and has attracted media attention. It has so far obtained a higher Altmetric attention score than 99% of its peers.

Context-aware stacked convolutional neural networks for classification of breast carcinomas in whole-slide histopathology images

In this paper, we present context-aware stacked convolutional neural networks (CNN) for classification of breast WSIs into normal/benign, ductal carcinoma in situ (DCIS), and invasive ductal carcinoma (IDC). The Preprint version of this paper can be found here: •

Deep learning-based assessment of tumor-associated stroma for diagnosing breast cancer in histopathology images

This is the first study showing the discriminating power of stromal properties of the tissue for diagnosing breast cancer. The corresponding manuscript accepted for publication in proceedings of IEEE ISBI can be found here •

Stain specific standardization of whole-slide histopathological images

Our paper presents a fully automated algorithm for standardization of whole-slide histopathological images to reduce the effect of these variations. The corresponding manuscript has been published in IEEE transactions on medical imaging 2016 (IEEE TMI) • Source code | Windows | Docker

Automated Detection of DCIS in Whole-Slide H&E Stained Breast Histopathology Images

This paper presents and evaluates a fully automatic method based on spatial clustering of superpixels for detection of ductal carcinoma in situ (DCIS) in digitized hematoxylin and eosin (H&E) stained histopathological slides of breast tissue. The corresponding manuscript has been published in IEEE transactions on medical imaging 2016 (IEEE TMI) •