Hierarchical Normalization for Robust Monocular Depth Estimation

Authors: Chi Zhang, Wei Yin, Billzb Wang, Gang Yu, BIN FU, Chunhua Shen

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our extensive experiments show that the proposed normalization strategy remarkably outperforms previous normalization methods, and we set new state-of-the-art on five zero-shot transfer benchmark datasets.
Researcher Affiliation Collaboration Chi Zhang1, Wei Yin2, Zhibin Wang1, Gang Yu1 , Bin Fu1, Chunhua Shen3 1Tecent PCG, China 2DJI Technology, China 3Zhejiang University, China
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide an explicit statement or link for open-source code availability for the described methodology.
Open Datasets Yes We follow Le Re S [37] to construct a mix-data training set, which includes 114K images from Taskonomy dataset, 121K images from DIML dataset, 48K images from Holopix50K, and 20K images from HRWSI [33].
Dataset Splits Yes We withhold 1K images from all datasets for validation during training.
Hardware Specification Yes The model is trained on 8 V100 GPU with the batch size of 32.
Software Dependencies No The paper mentions using "DPTHybrid [23]" but does not specify software versions for libraries, frameworks, or programming languages (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes We use the Adam optimizer with a learning rate of 10−5. The model is trained on 8 V100 GPU with the batch size of 32. In each mini-batch, we sample the equal number of images from different training data sources. For HDN in the spatial domain, we select the grid size Sspatial from {20, 21, 22, 23} to construct the hierarchical contexts. For HDN in depth domain, the group number Sdepth is chosen from {20, 21, 22}.