Knowledge-inspired 3D Scene Graph Prediction in Point Cloud

Authors: Shoulong Zhang, shuai li, Aimin Hao, Hong Qin

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experiments confirm that, our method can successfully learn representative knowledge embedding, and the obtained prior knowledge can effectively enhance the accuracy of relationship predictions.
Researcher Affiliation Academia Shoulong Zhang Beihang University shoulong.zhang@buaa.edu.cn Shuai Li Beihang University Peng Cheng Laboratory lishuai@buaa.edu.cn Aimin Hao Beihang University Peng Cheng Laboratory ham@buaa.edu.cn Hong Qin Stony Brook University (SUNY) qin@cs.stonybrook.edu
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes The code and instructions are included in the supplemental material.
Open Datasets Yes We train the meta-learning auto-encoder and the scene graph prediction model on the 3DSSG dataset [29] 1, a 3D scene graph dataset based on 3RScan [28].
Dataset Splits Yes With the same sub-scene split in [29], there are 3582 scenes in the training set and 548 for evaluation.
Hardware Specification Yes Our model is implemented in Py Torch. We trained our model on an Nvidia RTX 2080Ti GPU in a personal computer platform for 40 epochs with the ADAM optimizer.
Software Dependencies No The paper mentions 'Py Torch' as the implementation framework but does not specify its version or any other software dependencies with version numbers.
Experiment Setup Yes The initial learning rate is set to 0.0001, and the decay rate is 0.7 for every ten epochs. We followed the focal loss parameter settings in [29].