Attention-based Multi-Level Fusion Network for Light Field Depth Estimation
Authors: Jiaxin Chen, Shuo Zhang, Youfang Lin1009-1017
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show the proposed method achieves state-of-the-art performance in both quantitative and qualitative evaluation, which also ranks first in the commonly used HCI 4D Light Field Benchmark.In this section, we first introduce the detailed implementation of the experiments. Then the performance of our proposed method is compared with other state-of-the-art methods. Finally, we verify the effectiveness of the proposed attention module through ablation study. |
| Researcher Affiliation | Academia | Jiaxin Chen,1,2 Shuo Zhang,1,2,3 Youfang Lin1,2,3,4 1School of Computer and Information Technology, Beijing Jiaotong University, Beijing, China 2Beijing Key Laboratory of Traffic Data Analysis and Mining, Beijing, China 3CAAC Key Laboratory of Intelligent Passenger Service of Civil Aviation, Beijing, China 4Key Laboratory of Transport Industry of Big Data Appalication Technologies for Comprehensive Transport, Beijing, China |
| Pseudocode | No | No pseudocode or clearly labeled algorithm blocks were found in the paper. |
| Open Source Code | No | The paper does not provide any explicit statement or link indicating the availability of open-source code for the described methodology. |
| Open Datasets | Yes | We use the 4D synthetic LF Dataset (Honauer et al. 2016) in our experiment, in which images have 9 views and 512 512 spatial resolution. |
| Dataset Splits | Yes | Same with other networks (Shin et al. 2018; Yu-Ju et al. 2020), 16 images in Additional are used for training, 8 images in Stratified and Training for validating and 4 images in Test for testing. |
| Hardware Specification | Yes | The model is trained on an NVIDIA GTX 1080Ti GPU and takes about one week for training. |
| Software Dependencies | No | The paper states 'The tensorflow is used to implement the proposed network.' but does not provide specific version numbers for TensorFlow or any other software dependencies. |
| Experiment Setup | Yes | During training, patch-wise training is used by randomly cropping 32 32 gray-scale patches from the LF images.The L1 loss that measures the difference of estimated disparity bd and ground truth disparity dgt is used in our network. We use Adam optimizer (Kingma and Ba 2014) to optimize the network and set the batch size to 16. The learning rate is kept at 1e 3. |