A Blender extension for no-code generation of synthetic thermography and CV datasets
retained text
Hi! Some time ago I showcased here Rendersynth, a Blender extension to generate synthetic computer vision datasets without writing Blender Python scripts. I received some useful feedback so I thought I'd share a small update on the project: * Synthetic thermography: I've been working on IR-synth, a separate Blender add-on for generating synthetic thermal imagery from temperature fields and radiance models. * Pixel semantic segmentation: Replaced the previous approach with pixel-level segmentation masks with cryptomatte. * Quality of life improvements for the other formats (normal maps, depth maps, point clouds, keypoints, bounding boxes, ...) Source code for both modules is available at: For the main body: [https://github.com/lorenzozanizz/rendersynth](https://github.com/lorenzozanizz/rendersynth) For thermography: [https://github.com/lorenzozanizz/ir-synth](https://github.com/lorenzozanizz/ir-synth) The goal is still to make synthetic dataset generation accessible for smaller CV projects, without having to build and maintain custom Blender Python pipelines. I plan to integrate the two projects into a single suite and write hooks for external frameworks like Blenderproc. If you feel like trying it (Blender 4.5.0+) expect to find some nasty bugs, since its still a prototype. Any suggestions or criticism are welcome! Especially if you have any feedback or advice on the synthetic thermography component it would be greatly appreciated! Thank you for you attention! *As a side note, for the demonstration I used free models from SketchFab, 3DAI, CGTrader*
document fetched 2026-10-27 00:24 · canonical https://reddit.com/r/computervision/comments/1xfut6u/a_blender_extension_for_nocode_generation_of