SLIBO-Net: Floorplan Reconstruction via Slicing Box Representation with Local Geometry Regularization

Authors: Jheng-Wei Su, Kuei-Yu Tung, Chi-Han Peng, Peter Wonka, Hung-Kuo (James) Chu

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conducted experiments on a large-scale indoor synthetic dataset called Structured3D [25]. Following the approach in [9, 24], we divided the complete dataset into 2991 training samples, 250 validation samples, and 241 test samples
Researcher Affiliation Academia 1National Tsing Hua University 2National Yang Ming Chiao Tung University 3King Abdullah University of Science and Technology
Pseudocode No No pseudocode or algorithm blocks are explicitly provided in the paper.
Open Source Code Yes Our code and dataset are available online1. 1https://ericsujw.github.io/SLIBO-Net/
Open Datasets Yes We conducted experiments on a large-scale indoor synthetic dataset called Structured3D [25].
Dataset Splits Yes We divided the complete dataset into 2991 training samples, 250 validation samples, and 241 test samples
Hardware Specification Yes We implemented our model in Py Torch and trained our center transformer Tcenter on 2 NVIDIA V100s for 2 days. The box transformer Tbox was trained on 8 NVIDIA V100s for 1.9 days.
Software Dependencies No The paper mentions implementing the model in
Experiment Setup Yes We use the Adam optimizer with b1=0.9 and b2=0.999. The learning rates of two transformers and the Res Net-50 are 2.5e-4, and 1e-5, and we train the center transformer for 5000 epochs, box transformer for 16000 epochs with batch size 128 and 123 for each GPU. We empirically set 1 = 1, 2 = 1 in Equation 5, 3 = 1, 4 = 1, 5 = 5., 6 = 1 in Equation 10, = 2 in Equation 4, and M = 25, N = 50 in Section 3.3.