spatialreasoners/srmbench

Benchmarks for generative models from [ICML2025] "Spatial Reasoning with Denoising Models"

34
/ 100
Emerging

This project provides standardized tests to see how well image generation models understand complex spatial rules. It takes in images with specific challenges, like partially filled Sudoku grids or scenes with objects, and then measures if the generated images correctly follow spatial constraints. Researchers and practitioners in computer vision and generative AI would use this to rigorously benchmark and improve their image generation models.

Used by 1 other package. Available on PyPI.

Use this if you are developing or evaluating generative image models and need to assess their ability to reason about and adhere to intricate spatial relationships and constraints in generated content.

Not ideal if you are looking for a general-purpose dataset for training image generation models without a specific focus on evaluating spatial reasoning.

generative-ai-evaluation computer-vision-benchmarking spatial-reasoning-ai image-synthesis-testing ai-model-validation
Maintenance 6 / 25
Adoption 6 / 25
Maturity 22 / 25
Community 0 / 25

How are scores calculated?

Stars

9

Forks

Language

Python

License

MIT

Last pushed

Nov 11, 2025

Commits (30d)

0

Dependencies

8

Reverse dependents

1

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/generative-ai/spatialreasoners/srmbench"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.