Recent News

PicArrange 3.0 now finds images with words

PicArrange now supports color search

🎨 Find images with similar colors
📍 Search photos by location
🧭 Filter images using GPS data

Tutorial on Visual Information Retrieval

The slides and Jupyter notebooks of our ECIR2025 tutorial are available.

Winner of the SISAP Indexing Challange

Our adaption of the Exploration Graph to low recall regimes won us the SISAP Indexing Challange.

PicArrange 3.0 now finds images with words

Find your photos with words

PicArrange 3.0 allows to find your images with descriptions. Similarity search has been improved.

About Us Visual Computing Group @ HTW Berlin

About Us

The Visual Computing Group is a collaboration of various researchers at the HTW Berlin who work in the fields of:

  • Machine Learning
  • Information Retrieval
  • Visual clustering & sorting
  • Computer Vision
  • Immersive Imaging
  • Neural Rendering
More about us

Recent Publications

Dynamic Exploration Graph: A Novel Approach for Efficient Nearest Neighbor Search in Evolving Multimedia Datasets

Nico Hezel, Kai Uwe Barthel, Bruno Schilling, Konstantin Schall, Klaus Jung
31th International Conference on MultiMedia Modeling (MMM 2025)

The proposed Dynamic Exploration Graph (DEG) enhances Approximate Nearest Neighbor Search (ANNS) with efficient dynamic update support, outperforming existing methods on dynamic dataset while matching state-of-the-art static performance.

RegSegField: Mask-Regularization and Hierarchical Segmentation for Novel View Synthesis from Sparse Inputs

Kai Gu, Thomas Maugey, Sebastian Knorr, Christine Guillemot
21st ACM SIGGRAPH Conference on Visual Media Production (CVMP 2024)

We introduce RegSegField, a novel pipeline to utilize 2D segmentations to aid the reconstruction of objects and parts. This method introduces a novel mask-visibility loss by matching 2D segments across different views, thus defining the 3D regions for different objects.

Adapting the Exploration Graph for High Throughput in Low Recall Regimes

Nico Hezel, Bruno Schilling, Kai Uwe Barthel, Konstantin Schall, Klaus Jung
17th International Conference on Similarity Search and Applications (SISAP 2024)

We adapted our Exploration Graph for low-recall regimes and large-scale datasets (≥100 million data points) by integrating feature compression, path optimization, and smart entry points. These enhancements led to our success in the SISAP Indexing Challenge.

Optimizing CLIP Models for Image Retrieval with Maintained Joint-Embedding Alignment

Konstantin Schall, Kai Uwe Barthel, Nico Hezel, Klaus Jung
17th International Conference on Similarity Search and Applications, SISAP 2024

We propose two methods that significantly improve pre-trained CLIP models for image-to-image retrieval, while preserving the joint-embedding alignement and text-based task qualities. Our optimized models permit maintaining a single embedding per image, significantly simplifying the infrastructure needed for large-scale multi-modal similarity search systems.