ComPhy: Compositional Physical Reasoning of Objects and Events from Videos

Authors: Zhenfang Chen, Kexin Yi, Yunzhu Li, Mingyu Ding, Antonio Torralba, Joshua B. Tenenbaum, Chuang Gan

ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, We evaluate various baselines and analyze their results to study Com Phy thoroughly. We summarize the question-answering results of different baseline models in Table 2.
Researcher Affiliation Collaboration Zhenfang Chen MIT-IBM Watson AI Lab Kexin Yi Harvard University Yunzhu Li MIT Mingyu Ding The University of Hong Kong Antonio Torralba MIT Joshua B. Tenenbaum MIT BCS, CBMM, CSAIL Chuang Gan MIT-IBM Watson AI Lab
Pseudocode No The paper describes the model architecture and its components but does not include any pseudocode or algorithm blocks.
Open Source Code No Project page: https://comphyreasoning.github.io. The project page currently states 'Code will be released soon'.
Open Datasets Yes Project page: https://comphyreasoning.github.io (This link leads to the dataset download page, indicating public availability of the Com Phy dataset used for training, validation, and testing.)
Dataset Splits Yes Overall, Com Phy has 8,000 sets for training, 2,000 sets for validation, and 2,000 for testing.
Hardware Specification Yes We train the all the modules using Pytorch library Paszke et al. (2017) on Titan Nvidia GTX 1080-Ti GPUs.
Software Dependencies No The paper mentions using 'Pytorch library' and 'Mask-RCNN' but does not specify their version numbers or other key software dependencies with specific versions.
Experiment Setup No The paper describes the training objectives and data usage for different modules (e.g., cross-entropy loss, mean square error loss) but does not provide specific hyperparameter values like learning rate, batch size, or number of epochs.