Implicit Modeling of Non-rigid Objects with Cross-Category Signals

Authors: Yuchun Liu, Benjamin Planche, Meng Zheng, Zhongpai Gao, Pierre Sibut-Bourde, Fan Yang, Terrence Chen, Ziyan Wu

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results illustrate that our model can proficiently learn the shape representation of each organ and their relations to others, to the point that shapes missing from unseen instances can be consistently recovered by our method. Finally, MODIF can also propagate semantic information throughout the population via accurate point correspondences. We evaluate our solution on 3 different datasets over multiple tasks and show that MODIF consistently outperforms state-of-the-art methods, even when the latter are provided with additional supervision.
Researcher Affiliation Industry United Imaging Intelligence {yuchun.liu01, benjamin.planche, meng.zheng, zhongpai.gao, fan.yang03, terrence.chen, ziyan.wu}@uii-ai.com
Pseudocode No The paper does not contain any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain an explicit statement about the release of source code or a link to a code repository.
Open Datasets Yes Since our model is focused on modeling sets of non-rigid objects, we opt for three medical shape benchmarks: WORD (Luo et al. 2022), Abdomen CT (Ma et al. 2022) and Multi-Modality Whole-Heart Segmentation (MMWHS) (Zhuang and Shen 2016).
Dataset Splits No The paper mentions '30 training / 10 testing samples for MMWHS, 100/20 for WORD, and 37/10 for Abdomen CT', clearly specifying training and testing splits. However, it does not explicitly mention a 'validation' split or its size for the datasets used in the experiments.
Hardware Specification Yes Our model is trained on three NVIDIA RTX A40 GPUs for 300 epochs ( 1.5 hours over the WORD dataset).
Software Dependencies No The paper mentions using 'MLP architectures as in DIF-Net (Deng, Yang, and Tong 2021)' but does not specify exact version numbers for programming languages, libraries, or frameworks (e.g., Python, PyTorch, TensorFlow, CUDA).
Experiment Setup Yes We fix the dimensionality of per-object codes αi,j to 128, and the value of λC to 102. Other hyper-parameters are listed in annex. Our model is trained on three NVIDIA RTX A40 GPUs for 300 epochs ( 1.5 hours over the WORD dataset).