Selected Publications
AVisT: A Benchmark for Visual Object Tracking in Adverse Visibility
AVisT is a new video object tracking dataset containing challenging video sequences in adverse visibility scenarios.
arXiv, 2022
Robust Visual Tracking by Segmentation
RTS is a segmentation-centric tracker that uses segmentation masks as internal target representation in order to produce accurate segmentation masks of the target instead of bounding boxes in each frame.
ECCV, 2022
Transforming Model Prediction for Tracking
ToMP is a novel DCF based tracker that uses a Transformer based model predictor for both target localization and bounding box estimation.
CVPR, 2022
Learning Target Candidate Association to Keep Track of What Not to Track
Novel tracking method that tracks both the target and distractor object using a target candidate association network.
ICCV, 2021
Towards Closing the Gap in Weakly Supervised Semantic Segmentation with DCNNs: Combining Local and Global Models
Training strategy to reduced the gap between a network trained on full or on weak annotations.
CVIU, 2021
Adversarial Feature Distribution Alignment for Semi-Supervised Learning
AFDA is a semi-supervised learning method that aligns feature distributions between labeled and unlabeled to boost the classification accuracy.
CVIU, 2021
Group Sparsity: The Hinge Between Filter Pruning and Decomposition for Network Compression
Novel network compression techique that unifies filter pruning and low-rank decomposition employing group sparsity regularization.
CVPR, 2020
Efficient Video Semantic Segmentation with Labels Propagation and Refinement
Efficient Video Segmentation (EVS) pipeline that combines a fast CPU optical flow method with two CNNs to predict dense semantic labels.
WACV, 2020
Adversarial Sampling for Active Learning
ASAL is a new active learning strategy that synthesizes uncertain samples instead of performing an exhaustive search in each active learning cycle.
WACV, 2020
Origami: A Convolutional Network Accelerator
Convolutional network accelerator which is able to handle GPU-CNN sizes within the power envelope of embedded systems.
GLSVLSI, 2015
