UncNeRF: Uncovering Heavily Occluded Object with Multi-view Clues

This repository contains the code implementation for UncNeRF: Uncovering Heavily Occluded Object with Multi-view Clues.

⚙️ Environment Setup

1. Install PyTorch & CUDA Ensure you have installed a PyTorch version that matches your local CUDA environment. Please refer to the PyTorch official website for installation commands.

2. Install Python Dependencies Clone this repository and install the required packages:

pip install -r requirement.txt

3. Install Segment Anything Model (SAM) Our method requires SAM for specific processing steps. Please follow the official installation guide to set it up: facebookresearch/segment-anything

🗂️ Dataset

The custom dataset used in this project is publicly available on Hugging Face.

• Download Link: jiawei919/UncNeRF_dataset

Please download the dataset and place it in the appropriate data directory before running the experiments.

🚀 Quick Start

Training

To run the baseline model:

python cycle_train.py --config ./config/llff/xxx/xxx.txt

To run our proposed method (UncNeRF):

python pseudo_train.py --config ./config/llff/xxx/xxx.txt

Note: The rendering process is automatically integrated and executed during the training phase.

Rendering Only

If you already have a trained model and only want to render the novel views, run the following command:

python pseudo_train.py --render_only --config ./config/llff/xxx/xxx.txt