SyncDreamer: Generating Multiview-consistent Images from a Single-view Image
Authors: Yuan Liu, Cheng Lin, Zijiao Zeng, Xiaoxiao Long, Lingjie Liu, Taku Komura, Wenping Wang
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4 EXPERIMENTS. We quantitatively compare Sync Dreamer with baseline methods on the Google Scanned Object (Downs et al., 2022) dataset. The results show that, in comparison with baseline methods, Sync Dreamer is able to generate more consistent images and reconstruct better shapes from input single-view images. |
| Researcher Affiliation | Collaboration | 1The university of Hong Kong 2Tencent Games 3University of Pennsylvania 4Texas A&M University |
| Pseudocode | No | The paper describes its methodology in Section 3 and provides a pipeline diagram in Figure 2, but it does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The abstract provides a “Project page: https://liuyuan-pal.github.io/Sync Dreamer/”. However, the paper does not contain an explicit statement that the source code for the described methodology is released or available, nor does the provided link directly point to a source code repository containing the code. |
| Open Datasets | Yes | We train Sync Dreamer on the Objaverse (Deitke et al., 2023b) dataset which contains about 800k objects. we adopt the Google Scanned Object (Downs et al., 2022) dataset as the evaluation dataset. |
| Dataset Splits | No | The paper mentions training on the Objaverse dataset and evaluating on the Google Scanned Object dataset, and uses a number of objects for quantitative comparisons. However, it does not explicitly specify the training, validation, and test dataset splits (e.g., percentages or exact counts) for its main experiments. |
| Hardware Specification | Yes | We train the Sync Dreamer for 80k steps ( 4 days) with 8 40G A100 GPUs using a total batch size of 192. |
| Software Dependencies | No | The paper mentions using “Stable Diffusion” and “Zero123” as backbone models, and “Carve Kit” for foreground masks. However, it does not provide specific version numbers for these or any other software dependencies (e.g., Python, PyTorch, CUDA versions). |
| Experiment Setup | Yes | We train the Sync Dreamer for 80k steps ( 4 days) with 8 40G A100 GPUs using a total batch size of 192. The learning rate is annealed from 5e-4 to 1e-5. On each step, we sample 4096 rays and sample 128 points on each ray for training. |