Enhancing High-Resolution 3D Generation through Pixel-wise Gradient Clipping
Authors: Zijie Pan, Jiachen Lu, Xiatian Zhu, Li Zhang
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that PGC can serve as a generic integrative plug-in, consistently benefiting existing SDS and LDM-based 3D generative models, leading to significant improvements in high-resolution 3D texture synthesis. 5 EXPERIMENTS |
| Researcher Affiliation | Academia | Zijie Pan1, Jiachen Lu1, Xiatian Zhu2, Li Zhang1 1School of Data Science, Fudan University 2University of Surrey |
| Pseudocode | No | The paper does not contain any structured pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | Yes | https://fudan-zvg.github.io/PGC-3D/ |
| Open Datasets | Yes | For evaluation, we use ridge regression methods with the COCO dataset (Lin et al., 2014) to determine the optimal configuration. |
| Dataset Splits | No | The paper mentions using a "uniform setting" but does not specify exact percentages or sample counts for training, validation, or test splits. It does not reference predefined splits with citations for dataset partitioning. |
| Hardware Specification | Yes | Specifically, we optimize the same texture and/or signed distance function (SDF) fields as Chen et al. (2023b) for 1200 iterations on two A6000 GPUs with batch size 4 by using Adam optimizer without weight decay. |
| Software Dependencies | No | The paper mentions software components like "Adam optimizer", "Stable Diffusion", "SDXL", "VAE", and "U-net" but does not provide specific version numbers for these, or for general programming environments like Python or PyTorch. |
| Experiment Setup | Yes | For all the experiments, we adopt the uniform setting without any hyperparameter tuning. Specifically, we optimize the same texture and/or signed distance function (SDF) fields as Chen et al. (2023b) for 1200 iterations on two A6000 GPUs with batch size 4 by using Adam optimizer without weight decay. The learning rates are set to constant 1 10 3 for texture field and 1 10 5 for SDF field. For the sampling, we set the initial mesh normalized in [ 0.8, 0.8]3, focal range [0.7, 1.35], radius range [2.0, 2.5], elevation range [ 10 , 45 ] and azimuth angle range [0 , 360 ]. In SDS, we set CFG 100, t U(0.02, 0.5) and w(t) = σ2 t . In PGC, we use PGC-N for PGC as default and set the threshold c = 0.1. The clipping threshold is studied in Section A.3. |