In the Visual Intelligence Lab (
), we work with visual data targeting for humanlike perception, understanding and creation of visual worlds. The research in involves both discriminative tasks in image classification, object detection, semantic segmentation as well as generative tasks in image translation, image composition, and image editing. We work with different data modalities including 2D optical images, multi-spectral images, 3D point cloud, etc., and interpret them with different learning approaches including supervised learning, semi-supervised learning, transfer learning (domain adaption and generalization), self-supervised learning, etc. Altogether, VILab works towards humanlike visual intelligence.
Latest News
- Open-Vocabulary Object Detection via Language Hierarchy
- Historical Test-time Prompt Tuning for Vision Foundation Models
- Mitigating Object Hallucination via Concentric Causal Attention
- Domain Adaptation for Large-Vocabulary Object Detectors
- MonoMAE: Enhancing Monocular 3D Detection through Depth-Aware Masked Autoencoders
- Masked AutoDecoder is Effective Multi-Task Vision Generalist
- FreGS: 3D Gaussian Splatting with Progressive Frequency Regularization
- Weakly Supervised Monocular 3D Detection with a Single-View Image
- Cross-Domain Few-Shot Segmentation via Iterative Support-Query Correspondence Mining
- Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding
- Efficient Test-Time Adaptation of Vision-Language Models
- 3D Semantic Segmentation in the Wild: Learning Generalized Models for Adverse-Condition Point Clouds
- KD-DLGAN: Data Limited Image Generation via Knowledge Distillation
- StyleRF: Zero-shot 3D Style Transfer of Neural Radiance Fields
- Hierarchical Mask Calibration for Unified Domain Adaptive Panoptic Segmentation
- FAC: 3D Representation Learning via Foreground Aware Feature Contrast
- Regularized Vector Quantization for Tokenized Image Synthesis
- Backdoor Attacks Against Deep Image Compression via Adaptive Frequency Trigger
- Towards Efficient Use of Multi-Scale Features in Transformer-Based Object Detectors
- Da-detr: Domain adaptive detection transformer by hybrid attention
- Auto-regressive Image Synthesis with Joint Quantization [oral]
- Language Matters: A Weakly Supervised Vision-Language Pre-training Approach for Scene Text Detection and Spotting [oral]
- Bi-level feature alignment for versatile image translation and manipulation
- Contextual Text Block Detection towards Scene Text Understanding
- Domain Adaptive Video Segmentation via Temporal Pseudo Supervision
- Marginal Contrastive Correspondence for Guided Image Generation [oral]
- PTTR: Relational 3D Point Cloud Object Tracking with Transformer
- Modulated Contrast for Versatile Image Synthesis
- Spectral Unsupervised Domain Adaptation for Visual Recognition
- Category contrast for unsupervised domain adaptation in visual tasks
- Fourier Document Restoration for Robust Document Dewarping and Recognition
- Unbiased Subclass Regularization for Semi-Supervised Semantic Segmentation
- Accelerating DETR Convergence via Semantic-Aligned Matching
- WaveFill: A Wavelet-based Generation Network for Image Inpainting [oral]
- Sparse Needlets for Lighting Estimation with Spherical Transport Loss
- Domain Adaptive Video Segmentation via Temporal Consistency Regularization
- RDA: Robust Domain Adaptation via Fourier Adversarial Attacking
- Unsupervised domain adaptive 3d detection with multi-level consistency
- Skeleton cloud colorization for unsupervised 3d action representation learning
- Multiple Expert Brainstorming for Domain Adaptive Person Re-identification
- LEED: Label-Free Expression Editing via Disentanglement
- Contextual-Relation Consistent Domain Adaptation for Semantic Segmentation
- Collaborative Learning of Gesture Recognition and 3D Hand Pose Estimation with Multi-Order Feature Analysis [Spotlight]
- AMLN: Adversarial-based Mutual Learning Network for Online Knowledge Distillation