Paris AI, Machine Learning and Computer Vision Meetup

Jul 22, 2026
9:00 AM - 11:00 AM PST
Online. Register for the Zoom!

Contrastive learning on 3d point clouds for geometric defect detection

Reliable 3D defect detection in manufacturing is hard: the input is a point cloud — an unordered set that standard neural backbones cannot process directly; high-quality training data is scarce; and real scans are noisy and arrive in arbitrary orientations. We address these challenges in COSARAD, a contrastive learning framework that learns highly discriminative representations of object surface geometry under weak supervision.

When a test object arrives, we extract its features and compare them against a library of defect-free reference shapes for precise, interpretable defect localization — achieving state-of-the-art accuracy on industrial benchmarks such as Real3D-AD. In my talk, I'll cover the design choices behind the system, why contrastive representation learning is the right fit for sparse 3D data, and open problems in scaling inspection to production.

About the Speaker

Alexander Tarvo is a researcher at the University of Washington's MACS Lab, where he works on computer vision with applications in robotics. He holds a PhD in Software Engineering from Brown University and previously held research and engineering roles at Google, Microsoft, and IBM Research. His current research focuses on 3D vision and reinforcement learning for industrial robotics.

A Semantic and Occlusion-Aware Gaussian Mixture Probability Hypothesis Density Filter

Reliable and resilient multi-target tracking is foundational for safe autonomous driving, yet most perception pipelines frequently struggle with sensor noise, heavy clutter, and severe environmental occlusions. To resolve these limitations, this talk presents a novel Semantic-Occlusion Aware (S-OA) Gaussian Mixture Probability Hypothesis Density (GM-PHD) filter.

By combining geometric occlusion reasoning with deep learning-derived environmental semantics, the proposed framework adaptively initializes target tracking in regions where new targets are likely to appear. Evaluations demonstrate that this context-aware tracking system minimizes track initiation latency and preserves high tracking precision even under intense clutter.

Ultimately, this work demonstrates how embedding spatial and semantic structure into filtering yields a significantly more robust and resilient perception stack for autonomous navigation.

About the Speaker

Jovan Menezes is a PhD student at Cornell University, advised by Prof. Mark Campbell. His research focuses on developing scalable and resilient perception algorithms for autonomous driving. By leveraging concepts from probabilistic estimation and deep learning-based computer vision, the goal is to enable autonomous vehicles to perceive and navigate in challenging environments.

An Annotation-to-Detection Framework for Autonomous and Robust Vine Trunk Localization in the Field by Mobile Agricultural Robots

Autonomous robots struggle to detect objects in unstructured fields, requiring in-domain tuning with laborious manual data collection. In this work, we introduce a comprehensive annotation-to-detection framework designed to train a robust multi-modal detector using limited and partially labeled training data.

Our method combines cross-modal annotation transfer, early sensor fusion, and a multi-stage detection architecture to train and enhance multi-modal detection. Validated on vineyard trunk detection and paired with a custom LOAM algorithm, it localised over 70% of trees in one pass with under 0.37 m mean error.

Our system demonstrated that robust detection is achievable even with minimal initial annotations and human intervention.

About the Speaker

Dimitrios Chatziparaschis is a PhD candidate in EE, in University of California, Riverside. His main research lies at the intersection of computer vision, machine learning, and robotics. Main topics include 3D perception, multi-modal sensing, landmark detection, and localization in outdoor and dynamic settings.

vS-Graphs: Tightly Coupling Visual SLAM and 3D Scene Graphs Exploiting Hierarchical Scene Understanding

We introduce vS-Graphs, a novel real-time VSLAM framework that integrates vision-based scene understanding with map reconstruction and comprehensible graph-based representation. The framework infers structural elements (i.e., rooms and floors) from detected building components (i.e., walls and ground surfaces) and incorporates them into optimizable 3D scene graphs.

This solution enhances the reconstructed map's semantic richness, comprehensibility, and localization accuracy.

About the Speaker

Ali Tourani an R&D Specialist and a Senior Software Engineer with 8+ years of experience in practical computer vision and AI system design and deployment. Currently, he holds a Postdoctoral Research Associate position at the University of Luxembourg, where he develops vision-language models and generative AI solutions for real-world robotic applications.

Jul 23, 2026
9:00 AM - 11:00 AM PST
Online. Register for the Zoom!

Generative AI for Video Trailer Synthesis: From Extractive Heuristics to Autoregressive Creativity

The domain of automatic video trailer generation is currently undergoing a profound paradigm shift, transitioning from heuristicbased extraction methods to deep generative synthesis. While early methodologies relied heavily on low-level feature engineering, visual saliency, and rule-based heuristics to select representative shots, recent advancements in Large Language Models (LLMs), Multimodal Large Language Models (MLLMs), and diffusion-based video synthesis have enabled systems that not only identify key moments but also construct coherent, emotionally resonant narratives.

This survey provides a comprehensive technical review of this evolution, with a specific focus on generative techniques including autoregressive Transformers, LLM-orchestrated pipelines, and text-to-video foundation models like OpenAI's Sora and Google's Veo. We analyze the architectural progression from Graph Convolutional Networks (GCNs) to Trailer Generation Transformers (TGT), evaluate the economic implications of automated content velocity on User-Generated Content (UGC) platforms, and discuss the ethical challenges posed by high-fidelity neural synthesis.

By synthesizing insights from recent literature, this report establishes a new taxonomy for AI-driven trailer generation in the era of foundation models, suggesting that future promotional video systems will move beyond extractive selection toward controllable generative editing and semantic reconstruction of trailers.

About the Speaker

Abhishek Dharmaratnakar is an Engineering Leader at Google leading YouTube Premium. His work focuses on the intersection of hyperscale media infrastructure and generative artificial intelligence, directing cross-functional engineering organizations to redefine how billions of users consume and create content

Making Agent Systems Observable, Reliable, and Testable

In this talk, I’ll share practical lessons from building real agent systems in computer vision workflows, focusing on how to design evaluation loops, observability pipelines, and sandboxed environments that make agents reliable in practice. We’ll explore how to measure behavior end-to-end, test components independently, and build feedback loops that help agents improve over time, even as tools, models, and pipelines evolve. This talk is for engineers and builders who want to move beyond demos and learn how to make agent systems production-ready.

About the Speaker

Adonai Vera - Machine Learning Engineer & DevRel at Voxel51. With over 7 years of experience building computer vision and machine learning models using TensorFlow, Docker, and OpenCV.

Training-Free Object and Associated Effect Removal in Videos

I will be presenting our recent work, Object-WIPER, which focuses on removing objects and their associated effects from videos. Instead of fine-tuning models for each editing task, our method reuses the priors of pre-trained text-to-video models to perform object and effect removal in a training-free manner. We also curate a real world associated-effect benchmark and evaluation metric for more realistic assessment of video object removal.

About the Speaker

Saksham Singh Kushwaha is a candidate at UT Dallas, with research interests in audio-visual learning, spatial audio, and computer vision. I received my master’s degree from NYU and bachelor’s degree from IIT Delhi.

Turning Models into Systems: AI Architecture That Works

This talk explores what it really takes to make "intelligent systems" work in the messy, high-stakes reality of production environments – not just in demos or pilots. Most AI initiatives do not fail because the algorithms are weak, but because the surrounding system is not designed to handle uncertainty, change, and operational demands.

The session shows how to separate the concerns of building and improving models from their use in daily operations, and how to create a stable core of rules, safety, and business meaning around which smarter components can evolve.

Instead of treating AI as a magic add-on, the talk frames it as a capability that must be grounded in the organization's language, workflows, and responsibilities. It demonstrates how to design that core so that new models, tools, and data sources can be plugged in, compared, and replaced without breaking trust.

Attendees will leave with a clear mental model and a set of practical design ideas for turning clever prototypes into robust, understandable, and adaptable intelligent systems that people on the ground are willing to rely on.

About the Speaker

Dr. Nikita Golovko is a seasoned Solution Architect with over 16 years of experience in designing scalable, secure, and cost-effective software architectures for industrial and business-critical systems.

InstaDeep SAS
42 Rue de Paradis
75010 Paris, France

Finetuning VLMs for domain specific tasks

Vision-language models are becoming powerful general-purpose tools, but their real value often appears when they are adapted to domain-specific tasks. In this talk, I will present practical strategies for fine-tuning VLMs on specialized data, from dataset design and annotation quality to model selection, evaluation, and deployment constraints. I will discuss common challenges such as limited data, visual ambiguity, domain shift, hallucinations, and maintaining robustness in production. The session will focus on applied lessons for building reliable VLM systems that go beyond demos and solve concrete business or research problems.

About the Speaker

Amine Belhakimi is a Staff AI Engineer at GoPro with over 5 years of experience in computer vision and applied machine learning systems. His work focuses on bringing AI models from research to production, with a growing specialization in vision-language models, multimodal AI, and domain-specific AI applications.

Computer Vision at Nanoscale - Detecting, Segmenting and Analyzing Nanoparticles in microscopic images

This talk covers detecting, segmenting, and analyzing nanoparticles in microscopic images — including YOLO models, Classical CV, and Gradio.

About the Speaker

Atif Anwer is a computer vision and AI researcher with a passion for building intelligent systems that bridge robotics, deep learning, and real-world applications. He earned a joint Ph.D. from INSA Rouen Normandie and Universiti Teknologi PETRONAS, where his work focused on generative AI and image enhancement using polarimetric imaging data.

Towards a Resolution- and Modality-Agnostic Transformers for Earth Observation

Vision Transformers (ViT) dominate computer vision. However, their reliance on rigid patch projectors hinders transfer to Earth Observation (EO), where inputs vary widely in modality, scale, and resolution. We introduce UniverSat, a ViT-style backbone built around a Universal Patch Encoder that maps patches from arbitrary spatial, spectral, and temporal resolutions, and from both optical and non-optical sensors, into a shared embedding space with a single set of weights. This enables training one model on heterogeneous multimodal corpora in self-supervision, yielding robust sensor-agnostic spatial features.

About the Speaker

Guillaume Astruc is pursuing a PhD at Imagine Institute from École des Ponts ParisTech (ENPC), Lastig laboratory at National Institute of Geographic and Forest Information (IGN), and French National Centre for Space Studies (CNES).

Building Real-World Computer Vision Systems with Voxel51

This talk will explore practical workflows for building, evaluating, and improving modern computer vision systems. We’ll dive into real-world approaches to dataset curation, model analysis, multimodal AI workflows, and production-ready vision pipelines using open-source technologies.

The session is designed for engineers, researchers, and AI practitioners looking to better understand how teams are developing and scaling computer vision applications today. Expect practical demos, technical insights, and discussions around the evolving AI tooling ecosystem.

About the Speaker

Harpreet Sahota 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.

Efficient Image Generation through Smarter Data, Objectives, and Alignment

State-of-the-art image generation models require massive web scrapes and expensive post-training alignment. This talk explores three recent works that challenge the "bigger is better" paradigm to build efficient and controllable models. First, we show how ImageNet alone (only 1/1000th the training data of Stable Diffusion) can match billion-scale models using a fraction of the compute. Second, we introduce a frequency-balanced training objective that overcomes spectral bias, learning high-fidelity textures up to 40% faster. Finally, we present MIRO, a multi-reward pretraining method that bakes human preferences directly into the model, bypassing costly post-hoc RLHF and outperforming models 30x its size.

About the Speaker

Lucas Degeorge is a PhD Student at Ecole Polytechnique and Ecole des Ponts. His main research interest lies in multi-modal generative models.

Jul 29, 2026
9:00 AM - 11:00 AM PST
Online. Register for the Zoom!

The Agent Control Plane: Turning Coding Agents into Reliable Engineering Workflows

AI coding agents are powerful but often unreliable — they hallucinate, lose context, and produce inconsistent results across runs. In this talk, Alex introduces Atomic, an open-source control plane that adds persistent memory, deterministic workflow phases (Research → Specify → Implement → Ship), and human-in-the-loop gates around coding agents like Claude Code and GitHub Copilot. The result: repeatable, auditable engineering workflows that teams can actually trust in production.

About the Speaker

Alex Lavaee is an Applied AI engineer at Microsoft Research and the creator of Atomic, an open-source SDK that wraps deterministic, research-to-execution workflows around AI coding agents. He previously conducted AI research at Harvard Medical School and Boston University, and has worked as an MLE and data scientist at companies including Boeing and Themis AI, an MIT CSAIL spinoff.

UISurf: Toward Universal UI Automation with Cross-Environment Agents

In this talk, we introduce UISurf, an open-source multimodal agentic UI automation platform in which agents can perceive, reason, and collaborate across browser and desktop environments to complete end-to-end tasks that require interaction with multiple user interfaces.

UISurf comprises three main components: uisurf-agent, the runtime for UI automation agents; uisurf-admin, the session orchestration and management service; and uisurf-app, the full-stack user application. Its multi-agent architecture includes a planning_agent that transforms natural-language requests into structured execution plans, specialized Browser and Desktop Agents for environment-specific interaction, an automation_agent that coordinates execution and inter-agent handoff through Agent-to-Agent (A2A) communication, and a summarization_agent that produces the final task summary for the user. UISurf supports both autonomous execution and human-in-the-loop supervision, offering a practical and extensible framework for studying and deploying cross-environment UI automation.

About the Speaker

Dr. Henry Ruiz is a Research Scientist at Texas A&M University @ AgriLife Research, specializing in Artificial Intelligence (AI) and Remote Sensing. His work focuses on the development of advanced software systems and computational algorithms for analyzing multi-source remote sensing data, including satellite imagery, UAVs (Unmanned Aerial Vehicles), LiDAR (Light Detection and Ranging), and Ground Penetrating Radar (GPR).

From Manual Workflows to AI-Assisted Skills: Building Reliable Internal Automation

In this session, I will discuss how teams can turn repetitive manual workflows into reliable AI-assisted and automation-driven “skills.” I will share practical lessons from building internal tools for CAD and engineering workflows, including how automation can reduce manual effort, improve consistency, and support better process control. The talk will also cover why many AI/agent experiments fail when they are not connected to real team workflows, standards, and validation steps. Attendees will walk away with a practical framework for identifying repeatable workflows, designing useful internal tools, and adopting AI assistance without losing accuracy or trust.

About the Speaker

Janvi Vijaykumar Saddi - Janvi Saddi is a Computer Science graduate and CAD/Data Automation professional with experience in data center design workflows, AutoCAD automation, process improvement, and data analytics. She currently works as a CAD Tech 2 at Astreya, supporting Google data center design workflows by building internal tools that reduce manual effort, improve accuracy, and streamline engineering processes. Her background also includes SQL, Power BI, market research analytics, and AI-assisted development.

Building Safe Agent Sandboxes: Let Agents Act Without Breaking Production

AI agents become truly useful when they can take action, not just generate text. But giving agents access to code, data, and systems raises an important question: how do you let them explore, execute, fail, and improve without putting production at risk?

In this talk, we'll explore the sandbox pattern for agent systems and how to equip agents with tools to read, write, execute, and iterate within controlled environments while using permissions, human approval, and safety guardrails to keep them reliable. We'll cover practical architectures and lessons learned for building agents that can safely evolve from experimentation to production..

About the Speaker

Adonai Vera - Adonai Vera - Machine Learning Engineer & DevRel at Voxel51. With over 7 years of experience building computer vision and machine learning models using TensorFlow, Docker, and OpenCV.