Binocular-Guided 3D Gaussian Splatting with View Consistency for Sparse View Synthesis

Authors: Liang Han, Junsheng Zhou, Yu-Shen Liu, Zhizhong Han

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on the LLFF, DTU, and Blender datasets demonstrate that our method significantly outperforms the state-of-the-art methods.
Researcher Affiliation Academia School of Software, Tsinghua University, Beijing, China1, Department of Computer Science, Wayne State University, Detroit, USA2
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Project page is available at: https://hanl2010.github.io/Binocular3DGS/. The source code will be publicly available.
Open Datasets Yes We conduct experiments on three public datasets, including the LLFF dataset [27], the DTU dataset [18] and the Ne RF Blender Synthetic dataset (Blender) [28].
Dataset Splits Yes Following prior works [30, 54, 17], we used 3, 6, and 9 views as training sets for the LLFF and DTU datasets, and 8 images for training on the Blender dataset. The supplementary materials include detailed information on data splits.
Hardware Specification Yes We train our model on an RTX 3090 GPU
Software Dependencies No The paper mentions software like PDC-Net+ and LoFTR, but does not provide specific version numbers for general software dependencies (e.g., Python, PyTorch versions).
Experiment Setup Yes For the LLFF and DTU datasets, we utilize a pre-trained matching network PDC-Net+ [42] to obtain keypoints from input images...we train the LLFF and DTU datasets for 30,000 iterations, while the Blender dataset is trained for 7,000 iterations...The maximum distance dmax for camera shift is set to 0.4, the opacity decay coefficient λ is set to 0.995, and the β in the loss function 7 is set to 0.2 as in the original 3DGS [19].