Advances in Weakly-Supervised Learning and Data Annotation

The field of computer vision is moving towards leveraging weakly-supervised learning and innovative data annotation techniques to improve model performance and reduce manual labeling efforts. Recent research has focused on developing methods that can learn from pseudo-labels, noisy labels, and limited annotations, enabling the training of accurate models with minimal human supervision. Notable papers have proposed frameworks for generating high-quality pseudo-labels, refining label assignments, and transferring knowledge across datasets. These advancements have significant implications for applications such as object detection, visual grounding, and animal pose estimation. Noteworthy papers include Weakly-Supervised Affordance Grounding Guided by Part-Level Semantic Priors, which achieved a breakthrough improvement in affordance learning, and D2AF, a robust annotation framework for visual grounding that overcomes dataset size limitations and enriches the quantity and diversity of referring expressions. Additionally, the paper Auto-Labeling Data for Object Detection presents a viable alternative to standard labeling by configuring previously-trained vision-language foundation models to generate application-specific pseudo ground truth labels.

Sources

Weakly-Supervised Affordance Grounding Guided by Part-Level Semantic Priors

D2AF: A Dual-Driven Annotation and Filtering Framework for Visual Grounding

Auto-Labeling Data for Object Detection

Towards Auto-Annotation from Annotation Guidelines: A Benchmark through 3D LiDAR Detection

Labelling Data with Unknown References

Pre-trained Vision-Language Models Assisted Noisy Partial Label Learning

Animal Pose Labeling Using General-Purpose Point Trackers

Diffusion Domain Teacher: Diffusion Guided Domain Adaptive Object Detector

Noise-Resistant Label Reconstruction Feature Selection for Partial Multi-Label Learning

Using In-Context Learning for Automatic Defect Labelling of Display Manufacturing Data

Bridging Annotation Gaps: Transferring Labels to Align Object Detection Datasets

CzechLynx: A Dataset for Individual Identification and Pose Estimation of the Eurasian Lynx

Physical Annotation for Automated Optical Inspection: A Concept for In-Situ, Pointer-Based Trainingdata Generation

Tuning the Right Foundation Models is What you Need for Partial Label Learning

Do It Yourself: Learning Semantic Correspondence from Pseudo-Labels

Built with on top of