https://github.com/nepfaff/state_encoder_3d

3D Neural Scene Representations for View-Invariant State Representation Learning
https://github.com/nepfaff/state_encoder_3d

Last synced: 10 days ago
JSON representation

3D Neural Scene Representations for View-Invariant State Representation Learning

• Host: GitHub
• URL: https://github.com/nepfaff/state_encoder_3d
• Owner: nepfaff
• Created: 2023-04-23T18:14:41.000Z (almost 2 years ago)
• Default Branch: master
• Last Pushed: 2023-09-26T13:58:16.000Z (over 1 year ago)
• Last Synced: 2024-11-19T03:19:56.284Z (2 months ago)
• Language: Python
• Homepage:
• Size: 585 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# 3D Neural Scene Representations for View-Invariant State Representation Learning

My course project for MIT's [Advances in Computer Vision](http://6.8300.csail.mit.edu/sp23/) class.

## Installation

Clone the repo and execute the following commands from the repository's root.

Create a virtual environment:

```

python -m venv state_encoder_3d_env

```

Activate the environment:

```

source state_encoder_3d_env/bin/activate

```

Install the `state_encoder_3d` package in development mode:

```

pip install -e .

```

## Training and Evaluation

Training data can be generated using `scripts/generate_planar_cube_data.py`.

All training, evaluation, and visualization scripts can be found in the `scripts/` folder.

## Results

A detailed report can be found [here](./state_encoder_3d_report.pdf).

The below figure is taken from the report and compares the performance of different models.

![tSNE_neighbors_comparison](./tSNE_neighbors_comparison.png)