The local low-dimensionality of natural images
Authors: Olivier Henaff, Johannes Balle, Neil Rabinowitz, and Eero Simoncelli
ICLR 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show that images can be reconstructed nearly perfectly from estimates of the local filter response covariance alone, and with minimal degradation (either visual or MSE) from low-rank approximations of these covariances. As such, this representation holds much promise for use in applications such as denoising, compression, and texture representation, and may form a useful substrate for hierarchical decompositions. |
| Researcher Affiliation | Academia | Howard Hughes Medical Institute, and Center for Neural Science New York University New York, NY 10003, USA |
| Pseudocode | No | The paper does not contain any pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not mention providing open-source code for its described methodology, nor does it provide any links to a code repository. |
| Open Datasets | Yes | We trained the model described above on the van Hateren dataset (van Hateren & van der Schaaf, 1998) using the Torch machine learning library (Collobert et al., 2011). |
| Dataset Splits | No | The paper mentions using the van Hateren dataset for training and optimizing the filter bank, but it does not specify how the data was split into training, validation, or test sets, nor does it refer to predefined splits for reproduction. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments, such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | The paper mentions 'Torch machine learning library (Collobert et al., 2011)' but does not provide specific version numbers for Torch or any other ancillary software dependencies required for replication. |
| Experiment Setup | Yes | We used 20x20 pixel filter kernels, varying in number from 4 to 12, and estimated local dimensionality over neighborhoods of size 16x16 pixels, weighted by a Gaussian window with a standard deviation of 3 pixels. We fixed the blurring window h to be Gaussian with a standard deviation of 3 pixels... In the experiments below they were set to 3500 and 100 respectively. We optimized our filter bank using stochastic gradient descent with a fixed learning rate, chosen as high as possible without causing any instability. |