Description
REDVID-Gen is a research project exploring generative modelling as a high-level interface for particle-event simulation and detector-data generation. The aim is to let a user describe the desired event or dataset through a compact set of requirements, while a generative model produces the corresponding particle propagation and detector-hit data with minimal manual configuration.
Conventional simulation workflows often require users to select and configure many individual parameters before suitable data can be generated. REDVID-Gen investigates a complementary approach in which the user specifies the intended characteristics of the output and the generative model translates those requirements into coherent simulated events.
Generative event construction
The central research question is whether generative models can learn useful relationships between event-level requirements, particle properties, propagation behaviour, detector geometry, and the resulting hit patterns. Rather than generating isolated observations, the project targets structured event data in which trajectories and detector responses remain mutually consistent.
Depending on the selected modelling strategy, generation may operate directly on hit-level data or may first produce an intermediate event and propagation representation from which detector hits are derived. This allows the project to study the trade-off between direct generation and explicit physical or geometric structure.
Minimal user configuration
A key objective is to reduce the amount of detailed simulation configuration required from the user. Instead of exposing every low-level parameter as a mandatory input, REDVID-Gen aims to support concise specifications such as the desired event complexity, particle population, detector context, or dataset characteristics. The system can then infer or generate compatible lower-level details.
This interface is intended to make rapid dataset construction and exploratory studies easier while preserving control over the properties that matter for a particular experiment. Explicit parameters can still be retained where reproducibility, constraints, or targeted studies require them.
Relationship to REDVID
REDVID-Gen builds on the broader REDVID research direction of controllable, complexity-aware particle simulation and synthetic-data generation. REDVID provides an explicit simulation framework in which detector and event properties are configured directly. REDVID-Gen investigates how generative models can provide a higher-level layer that produces suitable event, propagation, and hit data from more compact user intent.
The two approaches are complementary: explicit reduced-order simulation provides transparent and controlled generation, while generative modelling may reduce configuration effort and enable richer distributions of events. Their combination can also provide a useful environment for validating generated samples against known simulation constraints.
Current status
REDVID-Gen is currently a research direction under development. Work is focused on defining useful input specifications, output representations, generative modelling strategies, and evaluation criteria for physical, geometric, and statistical consistency.
Outlook
The long-term goal is a flexible generative tool that can produce particle-event and detector-hit data for simulation studies and machine-learning experiments without requiring users to configure the full simulation process manually. This includes investigating how generated data can span controlled levels of event complexity while remaining suitable for reproducible scientific use.