Our paper has been accepted by BMVC 2021
Arvix version here
We propose a novel multimodal semi-supervised learning framework by introducing instance-level consistency constraint and a novel multimodal contrastive prototype (M2CP) loss. The instance-level consistency enforces the network to generate consistent representations for multimodal data of the same object regardless of its modality. The M2CP maintains a multimodal prototype for each class and learns features with small intra-class variations by minimizing the feature distance of each object to its prototype while maximizing the distance to the others. Our proposed framework significantly outperforms all the state-of-the-art counterparts for both classification and retrieval tasks by a large margin on the modelNet10 and ModelNet40 datasets.
Install Python
-- This repo is tested with Python 3.7.6.
Install NumPy
-- This repo is tested with NumPy 1.18.5. Please make sure your NumPy version is at least 1.18.
Install PyTorch
with CUDA -- This repo is tested with PyTorch 1.5.1, CUDA 10.2. It may work with newer versions, but that is not guaranteed. A lower version may be problematic.
Install TensorFlow
(for TensorBoard) -- This repo is tested with TensorFlow 2.2.0.
This netowrk is trained with two 16G Tesla V100 GPU
Remember change the dataloader root before training
python train.py
The trained model is save in checkpoints file. Remember change the test model name in test_12views.py
before testing.
python test_12views.py
We provide the pre-trained models of ModelNet40-10%
@article{chen2021multimodal,
title={Multimodal Semi-Supervised Learning for 3D Objects},
author={Chen, Zhimin and Jing, Longlong and Liang, Yang and Tian, YingLi and Li, Bing},
journal={arXiv preprint arXiv:2110.11601},
year={2021}
}