360-MLC: Multi-view Layout Consistency for Self-training and Hyper-parameter Tuning

Authors: Bolivar Solarte, Chin-Hsuan Wu, Yueh-Cheng Liu, Yi-Hsuan Tsai, Min Sun

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that our solution achieves favorable performance against state-of-the-art methods when self-training from three publicly available source datasets to a unique, newly labeled dataset consisting of multi-view images of the same scenes. In experiments, we leverage the MP3D-FPE [29] multi-view dataset as our unlabeled new domain.
Researcher Affiliation Collaboration Bolivar Solarte 1, Chin-Hsuan Wu 1, Yueh-Cheng Liu1, Yi-Hsuan Tsai 2, Min Sun1 1National Tsing Hua University, 2Phiar Technologies
Pseudocode No The paper does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No We will make our models, codes, and dataset available to the public. (This statement indicates future availability, not current concrete access.)
Open Datasets Yes We conduct extensive experiments using publicly available 360-image layout datasets: Matterport3D Floor Plan Estimation (MP3D-FPE) [29] as the target dataset, and three real-world datasets as the pre-training datasets, including Matterport Layout [36, 48], Zillow Indoor Dataset (ZIn D) [9], and the dataset used in Layout Net [47].
Dataset Splits No The paper specifies training and testing sets, but does not explicitly define a 'validation set' or 'validation split' with specific percentages or sample counts for hyperparameter tuning or model selection.
Hardware Specification Yes All models are trained on a single NVIDIA TITAN X GPU with 12 GB of memory.
Software Dependencies No The paper mentions 'Horizon Net [30] as our layout estimation backbone' but does not specify version numbers for any software dependencies like programming languages, libraries, or frameworks (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes We use the Adam optimizer to train the model for 300 epochs by setting the learning rate as 0.0001 and the batch size as 4.