Background
Before we can load model weights and run inference on static patches, we need to develop the piping around this model to load and and send input.
Description
Make it possible to run a model on a patch without any weights.
Create an interface for a model and define the input / output shape, you will need to do some research into the paper and pytorch to figure out what we are expecting input / output to look like.
Create two fake mock models that generate output, specifically a mock core model and a mock decoder.
The mock core model must:
- take in data of one shape and return data of two distinct shapes
- one is a low resolution elevation grid
- another is a latent map, a grid with multiple channels
- must be deterministic
The decoder model must:
- take a latent map and return a full resolution grid of a larger size
- must be deterministic
Both the low resolution grid and full resolution grid will eventually be passed into the elevation encoding.
Changes should live in src/terrain_diffusion/inference.py and tests/test_inference.py
Background
Before we can load model weights and run inference on static patches, we need to develop the piping around this model to load and and send input.
Description
Make it possible to run a model on a patch without any weights.
Create an interface for a model and define the input / output shape, you will need to do some research into the paper and pytorch to figure out what we are expecting input / output to look like.
Create two fake mock models that generate output, specifically a mock core model and a mock decoder.
The mock core model must:
The decoder model must:
Both the low resolution grid and full resolution grid will eventually be passed into the elevation encoding.
Changes should live in
src/terrain_diffusion/inference.pyandtests/test_inference.py