3D-Model-Reconstruction from 2D images

A Novel Hybrid Ensemble Approach For 3D Object
Reconstruction from Multi-View Monocular RGB images for Robotic Simulations

Ajith Balakrishnan, Dr. Sreeja S

implementation of paper - Refine3DNet: Scaling Precision in 3D Object Reconstruction from Multi-View RGB Images using Attention (https://arxiv.org/abs/2412.00731)

(1) Architecture

(2) STSO-JTSO Algorithm

(3) Sample Results

(4) Data

3D-R2N2 Dataset

https://github.com/chrischoy/3D-R2N2

LSM Dataset

https://github.com/akar43/lsm

(5) Released Model

Trained on 3D-R2N2 dataset

will release soon

(6) Requirements

python 3.5

tensorflow 1.11.0

numpy 1.13.3

scipy 0.19.0

matplotlib 2.0.2

skimage 0.13.0

(7) Run

Training

python main_AttSets.py

Test Demo (Download released model first)

python demo_AttSets.py

(8) Citation

If you use the paper or code for your research, please cite :

@article{Refine3DNet,
  title={Refine3DNet: Scaling Precision in 3D Object Reconstruction from Multi-View RGB Images using Attention},
  author={Ajith Balakrishnan, Dr Sreeja S, Dr Linu Shine},
  journal={ICVGIP},
  year={2024},
  doi={https://doi.org/10.1145/3702250.3702292}
}

(9) Special Thanks To

https://github.com/Yang7879/AttSets // https://github.com/chrischoy/3D-R2N2 // https://arxiv.org/pdf/1901.11153.pdf