ratvision

A Python library for simulating rat vision through 3D rendering.

ratvision provides a simple interface to render what a rat would see based on its position and head direction in a 3D environment. Three rendering back-ends are available: a Blender-based photorealistic renderer, a fast CPU raycaster, and a GPU-accelerated PyTorch renderer suitable for end-to-end differentiable pipelines.

Documentation

Full API documentation is available at marcoabrate.github.io/ratvision.

Installation

pip install ratvision

Or by cloning this repository:

git clone git@github.com:marcoabrate/ratvision.git
cd ratvision
pip install .

For GPU-accelerated rendering with TorchRenderer (Note: it installs torch):

pip install ratvision[gpu]

Quick start

Raycasting renderer (no external dependencies):

from ratvision import RaycastingRenderer

renderer = RaycastingRenderer()                    # default box environment
frame    = renderer.render_frame(0.3, 0.3, 0.0)   # (H, W) numpy array
frames   = renderer.render_path(positions, head_directions)

PyTorch renderer (GPU-accelerated):

from ratvision import TorchRenderer
import torch

renderer = TorchRenderer(config={'frame_dim': (64, 32)}).to('cuda')
frames   = renderer(positions_tensor, head_directions_tensor)  # (B, H, W) tensor

Blender renderer (photorealistic):

from ratvision import BlenderRenderer

renderer = BlenderRenderer(blender_exec='/path/to/blender')
renderer.render(positions, head_directions)

See the examples/ directory for full runnable demos with each back-end:

python examples/raycasting_render_demo.py
python examples/torch_render_demo.py
python examples/blender_render_demo.py --blender_exec "/path/to/Blender"

Requirements

• Python 3.9+
• Blender (only required for BlenderRenderer, not included in the package)

The Blender renderer was tested with Blender 3.6 on Linux and macOS.

Features

• Three rendering back-ends (Blender, CPU raycasting, GPU PyTorch)
• Generate rat-eye-view video animations from movement trajectories
• Easy to use Python API
• Compatible with custom 3D environments and procedural landmarks
• Built-in visualisation utilities (get_video_animation)
• GPU batch rendering for training loops (TorchRenderer)

Configuration options

Each renderer can be configured with parameters such as:

Additional Blender-specific options:

You can view and update the configuration at runtime:

renderer.config_description()          # print all available keys
renderer.update_config({'frame_dim': (64, 32)})

Customisable 3D environment

While ratvision comes with a default 3D environment, you can create custom environments programmatically or use your own Blender files:

from ratvision import BoxEnvironment, Landmark, RaycastingRenderer

env = BoxEnvironment(width=1.0, depth=1.0, height=0.8, wall_color=0.6)
renderer = RaycastingRenderer(env=env)

For Blender-based rendering with a custom .blend file:

renderer.update_config({'env_file': '/path/to/environment.blend'})

Note: All rendering and camera settings defined in the Blender file will be preserved. Only the parameters set through the config will be overwritten. For biologically-plausible rendering and camera settings, check the provided environment environments/box_messy.blend.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Author

Marco P Abrate marcopietro.abrate@gmail.com