Unsupervised object-centric video generation and decomposition in 3D

Authors: Paul Henderson, Christoph H. Lampert

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct detailed experiments on two datasets, going beyond the visual complexity supported by state-of-the-art generative approaches. We evaluate our method on depth-prediction and 3D object detection tasks which cannot be addressed by those earlier works and show it out-performs them even on 2D instance segmentation and tracking.
Researcher Affiliation Academia Paul Henderson IST Austria paul@pmh47.net Christoph H. Lampert IST Austria chl@ist.ac.at
Pseudocode No The paper describes the model architecture and training procedures textually and with diagrams, but it does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes All network architectures are given in the supplementary material, and code is available at https:// github.com/pmh47/o3v
Open Datasets No The paper states that for (ROOMS) 'we generated our own version of this dataset; details are given in the supplementary material' and for (TRAFFIC) 'is generated using the CARLA driving simulator [10]... We rendered a total of 5000 80-frame sequences'. While the base tools/simulators are cited, the paper does not provide a direct URL, DOI, or repository name for the specific datasets generated and used in their experiments.
Dataset Splits Yes We use two datasets in our experiments, reserving 10% of each for validation and 10% for testing.
Hardware Specification No We train each model on a single GPU, using gradient aggregation to increase the effective minibatch size. This mentions 'a single GPU' but does not specify the model or type of GPU, which is required for specific hardware details.
Software Dependencies No We implemented our model in Tensor Flow [1]. This mentions 'Tensor Flow' but does not provide a specific version number for it or any other key software dependencies.
Experiment Setup Yes Hyperparameters were set by grid searches over blocks of related parameters. The values used in our final experiments are given in the supplementary material.