Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session B: Wednesday, August 13, 1:00 – 4:00 pm, de Brug & E‑Hall

Controlled Synthetic Environments for Studying Mid-Level Vision in Artificial and Biological Systems

Joshua M. Martin¹, Thomas S. A. Wallis²; ¹Technische Universität Darmstadt, ²TU Darmstadt

Presenter: Joshua M. Martin

While artificial neural networks have advanced image-computable object recognition models, their ability to model mid-level vision remains limited. A key bottleneck is the scarcity of datasets with dense, high-quality annotations necessary for probing these intermediate computations. Here, we introduce a flexible synthetic image generation pipeline built in Blender that produces richly structured scenes with automatic pixel-level annotations, such as surface normals and segmentation masks. Drawing inspiration from digital embryos and dead leaves stimuli, the pipeline enables controlled manipulation of scene statistics and object properties. This provides targeted inputs for training and evaluating artificial neural networks, enabling detailed analysis of how specific visual features contribute to mid-level perceptual processes.

Topic Area: Visual Processing & Computational Vision

Extended Abstract: Full Text PDF