BMVC 2020

166 papers
The 2020 British Machine Vision Conference (BMVC) featured 208 papers covering a range of topics in computer vision and machine learning. Research themes focused heavily on data efficiency and model reliability, including active learning with variable annotation costs, calibrated data acquisition, and identifying diminishing returns in data labeling. Robustness was another central topic, with studies examining adversarial resilience, the limitations of obfuscated gradients, and generalization in few-shot learning. Advancements in deep metric learning and contrastive learning were also represented, specifically addressing geometric objectives and semantic collisions. Application-oriented research highlighted developments in video analysis and image retrieval. Notable works included reinforcement learning for video object segmentation, temporal modulation for thumbnail generation, and fine-grained sketch-based image retrieval. The proceedings also featured research into self-supervised anomaly detection and zero-shot learning via adversarial visual-semantic embeddings. Prolific authors contributing to the 2020 edition included Andrew Zisserman, Tinne Tuytelaars, and Yi-Zhe Song. These contributions collectively addressed challenges in balancing multiple learning objectives and improving cross-modal hierarchical modeling.

Top Research Directions

Research directions most represented by BMVC 2020 papers.

Reconstructive Regularization for Information Preservation in Latent Representations 2
Robust Feature Alignment for Unsupervised Domain Adaptation 2
Mitigating Input Perturbation Sensitivity in Deep Networks 2
Optimal Line Packing with Grassmannian Frames 1
The Mathematical Mechanisms of Dropout Regularization 1
Restoring Shift Invariance with Anti-Aliased Downsampling 1
The Performance Ceiling of Pre-trained Feature Dependency 1
Understanding the Joint Training Performance Gap 1
Scaling Metric Learning via Class Proxies 1
Inducing Sparsity in Regularized Optimal Transport 1
Point Cloud Registration via Continuous Probability Distributions 1
Overcoming Geometric Symmetry Priors in 3D Object Perception 1
Hardware and Computational Bottlenecks of Complex Activations 1
Overcoming Shift Variance in Deep Networks 1
Bridging the Discriminative to Generative Gap in Multimodal Learning 1

Top Authors

AuthorPapers
Andrew Zisserman3
Krystian Mikolajczyk3
Luc Van Gool3
Raja Giryes3
Wei Tang3
Yi-Hsuan Tsai3
Eugene Belilovsky2
Ko Nishino2
Le Wang2
Ming-Hsuan Yang2
Nathan Jacobs2
Rainer Stiefelhagen2
Theo Gevers2
Tinne Tuytelaars2
Vinay P. Namboodiri2
Vineeth N. Balasubramanian2
Xilin CHEN2
Xin Yu2
Yi Fang2
Yi-Zhe Song2

Impactful Papers

Papers from BMVC 2020 that shaped research clusters.

Foundational

Influential

Papers

Title
Unsupervised and Semi-supervised Novelty Detection using Variational Autoencoders in Opportunistic Science Missions
Unsupervised Domain Adaptation by Uncertain Feature Alignment
Unsupervised Domain Adaptation for Spatio-Temporal Action Localization
Video Region Annotation with Sparse Bounding Boxes
View-consistent 4D Light Field Depth Estimation
ViewSynth: Learning Local Features from Depth using View Synthesis
Visibility-aware Multi-view Stereo Network
Visualizing point cloud classifiers by curvature smoothing
WAMDA: Weighted Alignment of Sources for Multi-source Domain Adaptation
Weakly Paired Multi-Domain Image Translation
Weakly Supervised Generative Network for Multiple 3D Human Pose Hypotheses
Weakly-supervised Salient Instance Detection
What do CNNs gain by imitating the visual development of primate infants?
When Humans Meet Machines: Towards Efficient Segmentation Networks
Zero Shot Domain Generalization
Zero-Shot Domain Generalization

« prev Page 4/4