Learning the Geometry of Wave-Based Imaging
Authors: Konik Kothari, Maarten de Hoop, Ivan Dokmanić
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We showcase the advantages of learning geometry and the fact that the same network architecture can be applied to various problems. We choose three inverse problems as shown in Figure 4: reverse time continuation, inverse source problem, and reflector imaging. We train on 3000 samples of randomly oriented short thick box sources and test on samples from completely different distributions. For numerical results, please see Table 1 in Appendix A.2. |
| Researcher Affiliation | Academia | Konik Kothari UIUC kkothar3@illinois.edu Maarten de Hoop Rice University mvd2@rice.edu Ivan Dokmani c University of Basel ivan.dokmanic@unibas.ch |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks clearly labeled as such. |
| Open Source Code | Yes | Our codes are available2 for the community to reproduce our results and use our architectures for their own imaging modalities. 2https://github.com/kkothari93/fionet |
| Open Datasets | Yes | The dataset and training experiment details are given in Appendix D. We train on 3000 samples of randomly oriented short thick box sources and test on samples from completely different distributions. In Figure 5, we show the results on out-of-distribution data, such as Thick lines, Random shapes, Rotated MNIST, Exploding reflectors, and Celeb A face edges. Ziwei Liu et al. Deep Learning Face Attributes in the Wild . In: Proceedings of International Conference on Computer Vision (ICCV). Dec. 2015. |
| Dataset Splits | No | The paper mentions training and testing data but does not provide specific dataset split information for validation (e.g., percentages, sample counts, or explicit validation set mention). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions "Py Curvelab [47]" as a tool used but does not specify its version number or other software dependencies with their versions. |
| Experiment Setup | Yes | We devise a two-stage training strategy: first, we only train the geometry module... After e0 epochs, we train the full network as in (7) using the MSE loss. R := Rkmax is the maximum number of terms in (5) and is a hyperparameter in our training. The dataset and training experiment details are given in Appendix D. |