Shepard Convolutional Neural Networks
Authors: Jimmy SJ Ren, Li Xu, Qiong Yan, Wenxiu Sun
NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conducted experiments on two applications involving TVI: the inpainting and the super-resolution. The training data was generated by randomly sampling 1 million patches from 1000 natural images scraped from Flickr. Grayscale patches of size 48x48 were used for both tasks to facilitate the comparison with previous studies. All PSNR comparison in the experiment is based on grayscale results. Our model can be directly extended to process color images. |
| Researcher Affiliation | Industry | Jimmy SJ. Ren Sense Time Group Limited rensijie@sensetime.com Li Xu Sense Time Group Limited xuli@sensetime.com Qiong Yan Sense Time Group Limited yanqiong@sensetime.com Wenxiu Sun Sense Time Group Limited sunwenxiu@sensetime.com |
| Pseudocode | No | No pseudocode or algorithm blocks are present in the paper. |
| Open Source Code | No | The paper provides a 'Project page: http://www.deeplearning.cc/shepardcnn' but does not explicitly state that the source code for the methodology is available there or elsewhere. |
| Open Datasets | Yes | The training data was generated by randomly sampling 1 million patches from 1000 natural images scraped from Flickr. Grayscale patches of size 48x48 were used for both tasks to facilitate the comparison with previous studies. All PSNR comparison in the experiment is based on grayscale results. Our model can be directly extended to process color images. |
| Dataset Splits | No | The paper mentions 'training data' and 'test images' but does not specify a validation set or explicit numerical splits (percentages/counts) for any of the sets. |
| Hardware Specification | No | No specific hardware details (e.g., CPU/GPU models, cluster specifications) are mentioned in the paper regarding the experimental setup. |
| Software Dependencies | No | The paper mentions 'Ada Grad [17] was used in all experiments' but does not specify software names with version numbers for reproducibility. |
| Experiment Setup | Yes | The Sh CNN for inpainting is consists of five layers, two of which are Shepard interpolation layers. We use Re LU function [1] to impose nonlinearity in all our experiments. 4x4 filters were used in the first Shepard layer to generate 8 feature maps, followed by another Shepard interpolation layer with 4x4 filters. The rest of the Sh CNN is conventional CNN architecture. The filters for the third layer is with size 9x9x8, which are use to generate 128 feature maps. 1x1x128 filters are used in the fourth layer. 8x8 filters are used to carry out the reconstruction of image details. ... During training, weights were randomly initialized by drawing from a Gaussian distribution with zero mean and standard deviation of 0.03. Ada Grad [17] was used in all experiments with learning rate of 0.001 and fudge factor of 1e-6. |