Missed the European Conference on Computer Vision (ECCV) last month? Have no fear, we have collected some of the best research from the show into a series of online events.

**[Register for the Zoom](https://voxel51.com/computer-vision-events/eccv-2024-redux-day-1-nov-19-2024/)**

**Fast and Photo-realistic Novel View Synthesis from Sparse Images**

Novel view synthesis generates new perspectives of a scene from a set of 2D images, enabling 3D applications like VR/AR, robotics, and autonomous driving. Current state-of-the-art methods produce high-fidelity results but require a lot of images, while sparse-view approaches often suffer from artifacts or slow inference. In this talk, I will present my research work focused on developing fast and photorealistic novel view synthesis techniques capable of handling extremely sparse input views.

**ECCV 2024 Paper:** [CoherentGS: Sparse Novel View Synthesis with Coherent 3D Gaussians](https://www.ecva.net/papers/eccv_2024/papers_ECCV/papers/04306.pdf)

***About the Speaker***

[Avinash Paliwal ](https://www.linkedin.com/in/avinashpaliwal/)is a PhD Candidate in the Aggie Graphics Group at Texas A&M University. His research is focused on 3D Computer Vision and Computational Photography.

**Robust Calibration of Large Vision-Language Adapters**

We empirically demonstrate that popular CLIP adaptation approaches, such as Adapters, Prompt Learning, and Test-Time Adaptation, substantially degrade the calibration capabilities of the zero-shot baseline in the presence of distributional drift. We identify the increase in logit ranges as the underlying cause of miscalibration of CLIP adaptation methods, contrasting with previous work on calibrating fully-supervised models. Motivated by these observations, we present a simple and model-agnostic solution to mitigate miscalibration, by scaling the logit range of each sample to its zero-shot prediction logits

**ECCV 2024 Paper:** [Robust Calibration of Large Vision-Language Adapters](https://arxiv.org/abs/2407.13588)

***About the Speaker***

[Balamurali Murugesan](https://www.linkedin.com/in/balamurali-murugesan/) is currently pursuing his Ph.D. in developing reliable deep learning models. Earlier, he completed his master’s thesis on accelerating MRI reconstruction. He has published 25+ research articles in renowned venues.

**Tree-of-Life Meets AI: Knowledge-guided Generative Models for Understanding Species Evolution**

A central challenge in biology is understanding how organisms evolve and adapt to their environment, acquiring variations in observable traits across the tree of life. However, measuring these traits is often subjective and labor-intensive, making trait discovery a highly label-scarce problem. With the advent of large-scale biological image repositories and advances in generative modeling, there is now an opportunity to accelerate the discovery of evolutionary traits. This talk focuses on using generative models to visualize evolutionary changes directly from images without relying on trait labels.

**ECCV 2024 Paper:** [Hierarchical Conditioning of Diffusion Models Using Tree-of-Life for Studying Species Evolution](https://imageomics.github.io/phylo-diffusion/)

***About the Speaker***

[Mridul Khurana](https://www.linkedin.com/in/mridulk97/) is a PhD student at Virginia Tech and a researcher with the NSF Imageomics Institute. His research focuses on AI4Science, leveraging multimodal generative modeling to drive discoveries across scientific domains.

München AI Machine Learning and Computer Vision Meetup

Jimmy Guerrero

Patrick Madlindl

Advait Athreya 

ECCV Redux: Day 1 - Nov 19

Online event

**[Register for the Meetup](https://voxel51.com/computer-vision-events/london-ai-ml-amp-computer-vision-meetup-nov-19/)**

**Date, Time and Location**

*Date and Time*
Nov 19, 2024 from 5:30 PM to 8:30 PM

*Location*
The Meetup will take place at [Fora Space SOHO](https://www.foraspace.com/london-workspaces/broadwick-street) located at 33 Broadwick St in London.

**Understanding Memory in AI Agents and Agentic Systems**

Agentic Systems and extensible compound AI systems are revolutionizing LLM applications, positioning themselves as critical tools in modern AI development. These advanced systems go beyond traditional automation, offering capabilities that drive significant productivity and efficiency gains in enterprise and commercial workflows. However, adopting AI Agents and Agentic Systems at scale poses unique challenges, particularly in ensuring consistent performance, reliability, and scalability.

Central to overcoming these challenges is the role of memory. Memory within AI systems is not only essential for retaining operational data but also for enabling adaptive learning, entity profiling, and customized interactions. Different types of memory, such as short-term and long-term memory, play distinct roles in supporting an agent’s functionality. This talk will delve into the architecture of Agentic Systems and examine how various forms of memory—working memory, data stores, profilers, and toolboxes—contribute to creating robust, efficient, and scalable AI solutions. Attendees will gain insight into how memory is leveraged to enable learning from past executions, personalize interactions, and enhance system capabilities in complex AI applications.

*About the Speaker*

[Richmond Alake](https://www.linkedin.com/in/richmondalake/) is an AI/ML Developer Advocate at MongoDB, where he creates high-quality technical learning content for Developers and MongoDB customers building AI applications. In this role, he provides expert guidance on best practices for developing AI solutions that leverage Large Language Models (LLMs) and MongoDB, as well as offering insights on integrations and other critical aspects of AI development.

**How to Unlock More Value from Self-Driving Datasets**

AV/ADAS is one of the most advanced fields in Visual AI. However, getting your hands on a high quality dataset can be tough, let alone working with them to get a model to production. In this talk, I will show you the leading methods and tools to help visualize as well take these datasets to the next level. I will demonstrate how to clean and curate AV datasets as well as perform state of the art augmentations using diffusion models to create synthetic data that can empower the self driving car models of the future,

*About the Speaker*

[Daniel Gural](https://www.linkedin.com/in/daniel-gural/) is a seasoned Machine Learning Engineer at Voxel51 with a strong passion for empowering Data Scientists and ML Engineers to unlock the full potential of their data.

**Benchmarking and Optimization of LLMs**

In this session, we’ll focus on the critical role benchmarking plays in optimizing the use of large language models. We’ll dive into how to measure performance across different hardware setups, frameworks, and optimizations such as quantization and attention mechanisms.

*About the Speaker*

[Ciera Fowler](https://www.linkedin.com/in/daniel-gural/) is the ML Engineering Lead at [Ori](https://www.ori.co/), an AI native GPU cloud provider, and an MBA Student at London Business School. Ciera’s works on thought leadership pieces for Ori’s blog and speaking engagements focused on benchmarking and analysis of LLMs. She also posts tutorials and presents at tech meetups around London to help others start building their own LLM powered agents and applications.

London AI, Machine Learning and Computer Vision Meetup

Abhishek

Mukul Malhotra

Nov 19 - London AI, ML and Computer Vision Meetup

Fora Space SOHO

Join us for an evening of networking, talks, food and beverages at Macromedia University!

**Date and Time**

Nov 20, 2024 from 5:30 PM to 8:30 PM

**Location**

The Meetup will take place at [Macromedia University](https://www.macromedia-fachhochschule.de/en/locations/munich/), Sandstraße 9, Auditorium 608 in Munich

**Robust and Efficient Coupling of Perception to Actuation with Metric and Non-Metric Scene Representations**

We will present our work on robust and efficient ways, how camera information can be used to control an autonomous car. While many metric approaches require repeated intrinsic (sensor itself) and extrinsic (position of the sensor in the vehicle), our methods allow a direct control of the system solving the control problem directly in the retina (image plane) of the sensor. We will discuss the rich information encoded in the geometric projection of the scene that can be used for accurate navigation that includes uncertainty information necessary for fusion of diverse sensors and a novel way, how the dynamic information can be represented in the scene for planning in dynamic environments.

*About the Speaker*

[Prof. Burschka](https://www.professoren.tum.de/en/burschka-darius) of the Technical University of Munich conducts research into sensor systems in robotics and human-machine interfaces. Video-based navigation is one of his particular interests. This involves simulation of complex sensor systems through the analysis and fusion of sensor properties of physical sensor units and 3D reconstruction from the fusion of multimodal sensor data.

**Understanding Context in the Wild - AI Testing Automated Driving Systems**

Detailed contextual understanding is crucial for the testing of Automated Driving Systems (ADS). But, to provide high-quality and safe ADS, unseen events and potential weak points of the system need to be identified in the domain to be mitigated. This talk focuses on automated pipelines to identify the context, make driving data searchable and uncover potential weak points of driving systems automatically in the field.

*About the Speaker*

[Tin Stribor Sohn](https://www.linkedin.com/in/tin-stribor-sohn/) is a doctoral candidate and tech lead for vehicle data analytics at Porsche AG, dealing with scenario search and failure cause analysis for automated driving. Tin holds an MSc. in computer science and co-founder/CTO of an energy startup.

**Strategies Towards Reliable Scene Understanding for Autonomous Driving and Beyond**

Scene understanding is crucial for self-driving cars and autonomous agents to reliably perceive their surroundings in diverse, unpredictable conditions. This talk tackles these challenges by presenting a series of research papers that improve reliability through novel methods in 2D and 3D perception, focusing on robustness in extreme scenarios and generalization to unseen environments.

*About the Speaker*

[Stefano Gasperini](https://www.linkedin.com/in/stefano-gasperini/) is the Co-Founder & CEO of Visualais, a Computer Vision startup enabling the creation of 3D renderings from smartphone images. Beyond numerous top-tier research papers and outstanding reviewer awards from his PhD at TUM, as PostDoc, Stefano keeps contributing to Computer Vision and AI by advising a team of PhD students at TUM.

**How to Unlock More Value from Self-Driving Datasets**

AV/ADAS is one of the most advanced fields in Visual AI. However, getting your hands on a high quality dataset can be tough, let alone working with them to get a model to production. In this talk, I will show you the leading methods and tools to help visualize as well take these datasets to the next level. I will demonstrate how to clean and curate AV datasets as well as perform state of the art augmentations using diffusion models to create synthetic data that can empower the self driving car models of the future,

*About the Speaker*

[Daniel Gural](https://www.linkedin.com/in/daniel-gural/) is a seasoned Machine Learning Engineer at Voxel51 with a strong passion for empowering Data Scientists and ML Engineers to unlock the full potential of their data.

Giacomo Tomasi

Dirk Gómez

Nov 20 - Munich AI, ML and Computer Vision Meetup

 Macromedia University

**Pre-registering for the event is mandatory.**

**[Reserve your spot](https://voxel51.com/computer-vision-events/nov-20-san-francisco-ai-ml-amp-computer-vision-meetup/)**

**Date and Time**

Nov 20, 5:30 PM to 8:00 PM Pacific

**Location**

The Meetup will take place at [GitHub’s offices in San Francisco](https://www.google.com/maps/search/?api=1&query=GitHub&query_place_id=ChIJ32KaYX-AhYARNa93OffiHEk) located at 88 Colin P Kelly Jr St.

**Hands-On with Meta AI's CoTracker3: Parsing and Visualizing Point Tracking Output**

In this presentation we’ll explore Meta AI’s [CoTracker3](https://cotracker3.github.io/), a state-of-the-art point tracking model that effectively leverages real-world videos during training. He dives into the practical aspects of running inference with CoTracker3 and parsing its output into [FiftyOne](https://docs.voxel51.com/?_gl=1*13t0bw1*_gcl_au*NzQ3MzM1Mzg5LjE3MjYxNTQ3NzMuMTY5MDMwMTYzMy4xNzI5ODkzMDU0LjE3Mjk4OTM1NjU.), a powerful open-source tool for dataset curation, analysis, and visualization. Through a hands-on demonstration, Harpreet shows how to prepare a video for inference, run the model, examine its output, and parse the model’s output into FiftyOne’s keypoint format for seamless integration and visualization within the FiftyOne app.

*About the Speaker*

[Harpreet Sahota](https://www.linkedin.com/in/harpreetsahota204/) is a hacker-in-residence and machine learning engineer with a passion for deep learning and generative AI. He’s got a deep interest in RAG, Agents, and Multimodal AI.

**Why Speed Matters in Compound AI Systems: Making Models Go Vroom at Fireworks**

In this talk, we will analyze the critical role of speed in compound AI systems and how it impacts overall performance and user experience. Attendees will learn about some of the strategies we’ve implemented at [Fireworks](https://fireworks.ai/) to optimize model efficiency, reduce latency, and enhance responsiveness.

*About the Speaker*

[Mikiko Bazeley](https://www.linkedin.com/in/mikikobazeley/) is a Developer Relations Engineer at Fireworks.ai, specializing in MLOps and data science. With a passion for building high-performance AI systems, she collaborates with leading companies to drive innovation in generative AI. Mikiko loves empowering developers through hands-on workshops and engaging content, making complex AI concepts accessible and exciting.

**DIY LLMs**

In this talk, Charles will give a guided tour through the components of a self-hosted LLM service, from hardware considerations to engineering tools like ‘evals,’ all the way to the application layer. We’ll consider the open weights models, open source software, and infrastructure that power LLM applications. He will heavily shill the open source[ vLLM project](https://github.com/vllm-project/vllm).

*About the Speaker*

[Charles Frye](https://www.linkedin.com/in/charles-frye-38654abb/) builds applications of neural networks at Modal. He got his PhD at Berkeley for work on neural network optimization. He previously worked at Weights & Biases and Full Stack Deep Learning.

**From C Student to C-Suite without C++: AI Engineering for Entrepreneurs**

Building an 8-microservice LLM infrastructure with a team of three taught me that successful AI-assisted development isn’t about complex prompting – it’s about clear specifications and solid engineering practices. I’ll share how we use Cursor’s AI capabilities effectively: writing descriptive comments, maintaining clean architecture, and knowing when to break down complex tasks. By combining these fundamentals with well-documented frameworks like FastAPI and Pydantic, we’ve achieved 8-10x faster development while keeping code reliable. Learn from our experiences to separate what actually works from the AI hype.

*About the Speaker*

[Christos Magganas](https://www.linkedin.com/in/christos-magganas/) is an AI Engineer at AI Makerspace and a Founding Engineer at Toro, specializing in Python and AI/ML Ops. A firm believer in the collaborative spirit of innovation, he dedicates his time to community building and mentorship within the tech space. Through volunteering and consulting, Christos empowers others to explore the potential of AI/ML and contribute to solving real-world problems. He is driven by a vision of a future where technology serves humanity.

San Francisco AI and Machine Learning Meetup Group

Rajat Patel

Madeline

Henry Kendall

Nov 20 - SF AI, Machine Learning and Computer Vision Meetup

GitHub

Missed the European Conference on Computer Vision (ECCV) last month? Have no fear, we have collected some of the best research from the show into a series of online events.

**[Register for the Zoom](https://voxel51.com/computer-vision-events/eccv-2024-redux-day-3-nov-21-2024/)**

**Closing the Gap Between Satellite and Street-View Imagery Using Generative Models**

With the growing availability of satellite imagery (e.g., Google Earth), nearly every part of the world can be mapped, though street-view images remain limited. Creating street views from satellite data is crucial for applications like virtual model generation, media content enhancement, 3D gaming, and simulations. This task, known as satellite-to-ground cross-view synthesis, is tackled by our geometry-aware framework, which maintains geometric precision and relative geographical positioning using satellite information.

*ECCV 2024 Paper*

[Geospecific View Generation — Geometry-Context Aware High-resolution Ground View Inference from Satellite Views](https://arxiv.org/abs/2407.08061)

*About the Speaker*

[Ningli Xu](https://ninglixu.github.io/) is a Ph.D. student at The Ohio State University, specializing in generative AI and computer vision, with a focus on addressing image and video generation challenges in the geospatial domain.

**High-Efficiency 3D Scene Compression Using Self-Organizing Gaussians**

In just over a year, 3D Gaussian Splatting (3DGS) has made waves in computer vision for its remarkable speed, simplicity, and visual quality. Yet, even scenes of a single room can exceed a gigabyte in size, making it difficult to scale up to larger environments, like city blocks. In this talk, we’ll explore compression techniques to reduce the 3DGS memory footprint. We’ll dive deeply into our novel approach, Self-Organizing Gaussians, which proposes to map splatting attributes into a 2D grid, using a high-performance parallel linear assignment sorting developed to reorganize the splats on the fly. This grid assignment allows us to leverage traditional 2D image compression techniques like JPEG to efficiently store 3D data. Our method is quick and easy to decompress and provides a surprisingly competitive compression ratio. The drastically reduced memory requirements make this method perfect for efficiently streaming 3D scenes at large scales, which is especially useful for AR, VR and gaming applications.

*ECCV 2024 Paper*

[Compact 3D Scene Representation via Self-Organizing Gaussian Grids](https://fraunhoferhhi.github.io/Self-Organizing-Gaussians/)

*About the Speaker*

[Wieland Morgenstern](https://www.linkedin.com/in/wiemo/) is a Research Associate at the Computer Vision & Graphics group at Fraunhofer HHI and is pursuing a PhD at Humboldt University Berlin. His research focuses on representing 3D scenes and virtual humans.

**Skeleton Recall Loss for Connectivity Conserving and Resource Efficient Segmentation of Thin Tubular Structures**

We present Skeleton Recall Loss, a novel loss function for topologically accurate and efficient segmentation of thin, tubular structures, such as roads, nerves, or vessels. By circumventing expensive GPU-based operations, we reduce computational overheads by up to 90% compared to the current state-of-the-art, while achieving overall superior performance in segmentation accuracy and connectivity preservation. Additionally, it is the first multi-class capable loss function for thin structure segmentation.

*ECCV 2024 Paper*

[Skeleton Recall Loss for Connectivity Conserving and Resource Efficient Segmentation of Thin Tubular Structures](https://arxiv.org/abs/2404.03010)

About the Speakers

[Maximilian Rokuss](https://www.linkedin.com/in/maximilian-r-08997b176/) holds a M.Sc. in Physics from Heidelberg University, now PhD Student in Medical Image Computing at German Cancer Research Center (DKFZ) and Heidelberg University

[Yannick Kirchoff](https://www.linkedin.com/in/yannick-kirchhoff-518430201/) holds a M.Sc. in Physics from Heidelberg University, now PhD Student in Medical Image Computing at German Cancer Research Center (DKFZ) and Helmholtz Information and Data Science School for Health

Frank Merwerth 

flipsi schlotzi

ECCV Redux: Day 3 - Nov 21

AI, Machine Learning and Data Science Meetup Network

Cerca {networkName}

Gruppi

{PRO_NETWORK_NAME} fa parte di Meetup Pro con più di {PRO_MEMBER_COUNT} iscritti in {PRO_GROUP_COUNT} Meetup in tutto il mondo. Meetup Pro è lo strumento professionale per organizzare e comunicare in un network di utenti, partner, contributori e partecipanti.

This virtual group is for data scientists, machine learning engineers, and open source enthusiasts.

Every month we’ll bring you two diverse speakers working at the cutting edge of data science, machine learning, AI and computer vision. 

This Meetup is sponsored by Voxel51, the lead maintainers of the open source FiftyOne computer vision toolset.

AI, Machine Learning and Data Science Meetup Network

AI, Machine Learning and Data Science Meetup Network

Chi siamo

Cerca AI, Machine Learning and Data Science Meetup Network

Prossimi eventi

Gruppi