Multi-Resolution Monocular Depth Map Fusion by Self-Supervised Gradient-Based Composition
Authors: Yaqiao Dai, Renjiao Yi, Chenyang Zhu, Hongjun He, Kai Xu
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Quantitative evaluations demonstrate that the proposed method can be integrated into many fully convolutional monocular depth estimation backbones with a significant performance boost, leading to state-of-the-art results of detail enhancement on depth maps. In this section, we first introduce the benchmark datasets and metrics. Next, we compare with several state-of-the-art monocular depth estimation and refinement methods in aspects of several error metrics, robustness to noises, and running time. Lastly, we conduct several ablations to study the effectiveness of critical designs in the pipeline. |
| Researcher Affiliation | Academia | National University of Defense Technology chenyang.chandler.zhu, kevin.kai.xu@gmail.com |
| Pseudocode | No | No pseudocode or algorithm blocks were found. |
| Open Source Code | Yes | Codes are released at https://github.com/ yuinsky/gradient-based-depth-map-fusion. |
| Open Datasets | Yes | To evaluate the depth estimation ability of our method, we adopt several commonly used zero-shot datasets, which are Multiscopic (Yuan et al. 2021), Middlebury2021 (Scharstein et al. 2014) and Hypersim (Roberts et al. 2021). |
| Dataset Splits | No | No explicit training/validation/test dataset splits (e.g., percentages, sample counts, or specific predefined splits) were found for the datasets used in evaluation. |
| Hardware Specification | No | The paper does not explicitly describe the specific hardware (e.g., GPU model, CPU model, memory, or cloud instance type) used to run the experiments. |
| Software Dependencies | No | No specific software dependencies with version numbers (e.g., library names with specific versions) were mentioned in the paper. |
| Experiment Setup | Yes | All ablation alternatives are trained for 30 epochs, under identical training configurations. In guided filters, there are two parameters, the window size r is set to adjust the receptive field, along with an edge threshold ϵ. Here, we fix both parameters for the whole dataset and select high quality data as training set to get a reasonable guided filtered depth map. |