Self-Supervised Deep Learning on Point Clouds by Reconstructing Space

Authors: Jonathan Sauder, Bjarne Sievers

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show experimentally, that pre-training with our method before supervised training improves the performance of state-of-the-art models and significantly improves sample efficiency. 4 Experiments
Researcher Affiliation Academia Jonathan Sauder Hasso Plattner Institute Potsdam, Germany jonathan.sauder@student.hpi.de Bjarne Sievers Hasso Plattner Institute Potsdam, Germany bjarne.sievers@student.hpi.de
Pseudocode Yes Algorithm 1: Generation of Self-Supervised Labels
Open Source Code No The paper does not provide any explicit statements or links indicating that the source code for their methodology is open-source or publicly available.
Open Datasets Yes Model Net dataset [35], Shape Net dataset [5], Shape Net Part dataset [38], Stanford Large-Scale 3D Indoor Spaces (S3DIS) dataset [2]
Dataset Splits Yes For this we use the standard train/test split, with the same uniform point sample as defined in [23] with Model Net40 on 40 classes containing 9843 train and 2468 test models and Model Net10 on ten classes containing 3991 and 909 models respectively. We use the official train / validation / test splits [38].
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments (e.g., specific GPU or CPU models, memory, or cloud computing instance types).
Software Dependencies No The paper mentions various software components and models (e.g., PointNet, DGCNN, SVM, t-SNE, UMAP) but does not provide specific version numbers for any of them.
Experiment Setup Yes While k may be varied across domains... we list all results with k = 3. Additional details are discussed in Section 5. randomly rotating 15% of the individual voxels and randomly replacing one voxel in each input point cloud with a random voxel from a randomly drawn input point cloud from the same dataset leads to a slightly higher quality of the embeddings in the object classification task (consistently around 0.2% SVM accuracy in the downstream object classification task), therefore we kept this setup throughout all experiments. ...pre-training a DGCNN in a self-supervised manner on the Shape Net dataset with 1024 points chosen randomly from each model for 100 epochs before fully supervised training on the Model Net40 dataset.