Dynamic Spatial Propagation Network for Depth Completion
Authors: Yuankai Lin, Tao Cheng, Qi Zhong, Wending Zhou, Hua Yang1638-1646
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Dy SPN outperforms other state-of-the-art (So TA) methods on KITTI Depth Completion (DC) evaluation by the time of submission and is able to yield So TA performance in NYU Depth v2 dataset as well. |
| Researcher Affiliation | Academia | 1School of Mechanical Science and Enginering, Huazhong University of Science and Technology, China 2College of Urban Transportation and Logistics, Shenzhen Technology University, China 3Division of Logistics and Transportation, Tsinghua University, China 4School of Science and Engineering, The Chinese University of Hong Kong (Shenzhen), China |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. |
| Open Datasets | Yes | NYU Depth V2 dataset (Silberman et al. 2012) is comprised of RGB and depth images captured by a Microsoft Kinect camera in 464 indoor scenes. KITTI DC dataset (Uhrig et al. 2017) consists of over 90K RGB and Li DAR pairs for training, 1K pairs for validation, and another 1K pairs for testing. |
| Dataset Splits | Yes | Our model is trained on a subset of 50K images from the official training split and tested on the 654 images from the official labeled test set. KITTI DC dataset (Uhrig et al. 2017) consists of over 90K RGB and Li DAR pairs for training, 1K pairs for validation, and another 1K pairs for testing. |
| Hardware Specification | Yes | We implement our method on the Py Torch framework and train it with 4 NVIDIA RTX 2080 Ti GPUs. |
| Software Dependencies | No | The paper mentions 'Py Torch framework' but does not provide specific version numbers for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | All of our experiments are trained by an ADAM optimizer with β1=0.9, β2=0.999. The learning rate starts at 0.001 for the first 30 epochs and reduces to 0.002 with another 10 epochs of training. For the NYU Depth V2 dataset, the batch size is set to 24 and 500 depth pixels were randomly sampled from the ground truth depth. For the KITTI DC dataset, the batch size is 8. Besides, data augmentation techniques are used, including horizontal random flip and color jitter. |