dungeon-maps

Dungeon Maps

A tiny PyTorch library for depth map manipulations.

Version: 0.0.3a1

Features

	Batching	Multi-channels	GPU acceleration
Orthographic projection (Top-down map)	:heavy_check_mark:	:heavy_check_mark:	:heavy_check_mark:
Egocentric motion flow	:heavy_check_mark:	:heavy_check_mark:	:heavy_check_mark:
3D affine transformation (Camera space)	:heavy_check_mark:	:heavy_check_mark:	:heavy_check_mark:
2D affine transformation (Top-down view)	:heavy_check_mark:	:heavy_check_mark:	:heavy_check_mark:
Map builder	:heavy_check_mark:	:heavy_check_mark:	:heavy_check_mark:

Demos

Orthographic projection

Depth maps

(Watch this video in high quality)

This example shows how to project depth maps to top-down maps (plan view).

• Top left: RGB
• Top right: Depth map
• Bottom left: top-down maps in local space
• Bottom right: top-down maps in global space

Run this example

python -m dungeon_maps.demos.height_map.run

Control: W, A, S, D. Q for exit

Semantic segmentations

(Watch this video in height quality)

This example shows how to project arbitrary value maps, e.g. semantic segmentation, to top-down maps (plan view).

• Top left: RGB
• Top center: Depth map
• Top right: Semantic segmentation
• Bottom left: top-down maps in local space
• Bottom right: top-down maps in global space

Run this example

python -m dungeon_maps.demos.object_map.run

Control: W, A, S, D. Q for exit

Egocentric motion flow

(Watch this video in high quality)

This example shows how to calculate the flow fields caused by camera motion.

Run this example

python -m dungeon_maps.demos.ego_flow.run

Use case

One of the real case is to use this package along with the Habitat Lab to render the semantic topdown maps.

Installation

Basic requirements:

• Python >= 3.6
• PyTorch >= 1.8.0
• PyTorch Scatter

Install from pip

pip install dungeon_maps

Install from GitHub repo

pip install git+https://github.com.Ending2015a/dungeon_map.git@master

Install demos

pip install dungeon_maps[sim]