3DP3: 3D Scene Perception via Probabilistic Programming

Authors: Nishad Gothoskar, Marco Cusumano-Towner, Ben Zinberg, Matin Ghavamizadeh, Falk Pollok, Austin Garrett, Josh Tenenbaum, Dan Gutfreund, Vikash Mansinghka

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our results demonstrate that 3DP3 is more accurate at 6Do F object pose estimation from real images than deep learning baselines and shows better generalization to challenging scenes with novel viewpoints, contact, and partial observability.
Researcher Affiliation Collaboration 1MIT 2MIT-IBM Watson AI Lab {nishad,marcoct,bzinberg,mghavami,jbt,vkm}@mit.edu {falk.pollok,austin.garrett}@ibm.com dgutfre@us.ibm.com
Pseudocode No The paper describes algorithmic procedures (e.g., MCMC kernels) but does not present them in structured pseudocode blocks or clearly labeled algorithm sections in the main text.
Open Source Code Yes Our code is available at https://github.com/probcomp/Three DP3.
Open Datasets Yes We evaluate our scene graph inference algorithm on the YCB-Video [6] dataset consisting of real RGBD images and YCB-Challenging, our own synthetic dataset of scenes containing novel viewpoints, occlusions, and contact structure.
Dataset Splits No The paper mentions using the 'YCB-Video test set' for evaluation but does not specify the training, validation, or test split percentages or sample counts for dataset partitioning, nor does it provide access to custom splits.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., CPU/GPU models, memory specifications).
Software Dependencies No The paper states, 'Our model and inference algorithm are implemented in the Gen [11] probabilistic programming system,' but it does not provide specific version numbers for Gen or any other software libraries or dependencies.
Experiment Setup No The paper describes some aspects of the experimental setup, such as using Dense Fusion for pose initialization and details about proposal distributions for MCMC kernels, but it lacks specific hyperparameters (e.g., learning rate, batch size, number of epochs, optimizer settings) or system-level training configurations typically required for reproduction.