Puzzlefusion: Unleashing the Power of Diffusion Models for Spatial Puzzle Solving
Authors: Sepidehsadat (Sepid) Hossieni, Mohammad Amin Shabani, Saghar Irandoust, Yasutaka Furukawa
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The qualitative and quantitative evaluations demonstrate that our approach outperforms the competing methods by significant margins in all the tasks. |
| Researcher Affiliation | Academia | Sepidehsadat Hosseini, Mohammad Amin Shabani, Saghar Irandoust , Yasutaka Furukawa Simon Fraser University {sepidh, mshabani, sirandou, furukawa}@sfu.ca |
| Pseudocode | No | The paper describes its methods but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The abstract states 'We have provided code and data here.' and Section 1 states 'We will make all our code and data public.' However, no concrete link or specific repository is provided in the paper text. |
| Open Datasets | Yes | Magic Plan (https://www.magicplan.app), a mobile software company for real estate and construction, agrees to share production data with us and the research community, where the paper introduces the Magic Plan dataset, containing room shapes and their ground-truth arrangement for 98,780 single-story houses/apartments. We split the data into 93,780/5,000 training/testing samples. For the pictorial version, we have used images from COCO 2017 dataset Lin et al. (2014). |
| Dataset Splits | No | The paper specifies training and testing splits for its datasets (e.g., '93,780/5,000 training/testing samples' for Magic Plan), but does not explicitly mention a distinct validation split. |
| Hardware Specification | Yes | We have implemented the system with Py Torch Paszke et al. (2019), using a workstation with a 3.70GHz Intel i9-10900X CPU (20 cores) and two NVIDIA RTX A6000 GPUs. |
| Software Dependencies | No | The paper mentions 'Py Torch Paszke et al. (2019)' but does not provide a specific version number or other detailed software dependencies with version numbers. |
| Experiment Setup | Yes | We use the Adam W Loshchilov and Hutter (2017); Kingma and Ba (2014) optimizer with β1 = 0.9, β2 = 0.999, weight decay equal to 0.05, and the batch size of 512. The learning rate is initialized to 0.0005. |