Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
Saliency, Scale and Information: Towards a Unifying Theory
Authors: Shafin Rahman, Neil Bruce
NeurIPS 2015 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Based on the proposed definition of visual saliency, we demonstrate results competitive with the state-of-the art for both prediction of human fixations, and segmentation of salient objects. We also characterize different properties of this model including robustness to image transformations, and extension to a wide range of other data types with 3D mesh models serving as an example. Finally, we relate this proposal more generally to the role of saliency computation in visual information processing and draw connections to putative mechanisms for saliency computation in human vision. |
| Researcher Affiliation | Academia | Shafin Rahman Department of Computer Science University of Manitoba EMAIL Neil D.B. Bruce Department of Computer Science University of Manitoba EMAIL |
| Pseudocode | No | No pseudocode or clearly labeled algorithm blocks are present in the paper. |
| Open Source Code | No | The paper does not provide any concrete access information (e.g., repository links, explicit statements of code release) for the described methodology's source code. |
| Open Datasets | Yes | Benchmark results are provided for both fixation data and salient object segmentation. ... We have compared our results with several saliency and segmentation algorithms ... and across different datasets. ... To examine affine invariance, we have to used image samples from a classic benchmark [5] which represent changes in zoom+rotation, blur, lighting and viewpoint. |
| Dataset Splits | No | The paper utilizes standard benchmark datasets for evaluation but does not specify details regarding training, validation, or test splits for its own experimental setup. It does not mention cross-validation. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, memory specifications, cloud instance types) used for running the experiments are provided in the paper. |
| Software Dependencies | No | No specific software dependencies with version numbers (e.g., library names like PyTorch 1.9, or specific solver versions) are listed in the paper. |
| Experiment Setup | Yes | The size of Gaussian kernel Gσb determines the spatial scale. 25 different Kernel sizes are considered in a range from 3x3 to 125x125 pixels with the standard deviation σb equal to one third of the kernel width. The Gaussian kernel that defines color distance Gσi is determined by the standard deviation σi. We tested values for σi ranging from 0.1 to 10. For post processing standard bilateral filtering (BB), a kernel size of 9x9 is used, and center bias results are based on a fixed σcb = 5 for the kernel Gσcb for CB-1. For the second alternative method (CB-2) one Gaussian kernel Gσi is used with σi = 10. All of these settings have also considered different scaling factors applied to the overall image 0.25, 0.5 and 1 and in most cases, results corresponding to the resize factor of 0.25 are best. |