Compressing Images by Encoding Their Latent Representations with Relative Entropy Coding
Authors: Gergely Flamich, Marton Havasi, José Miguel Hernández-Lobato
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | supported by our empirical results obtained on the Cifar10, Image Net32 and Kodak datasets. Moreover, unlike previous bits-back methods, REC is immediately applicable to lossy compression, where it is competitive with the state-of-the-art on the Kodak dataset. Our experiments are implemented in Tensor Flow (Abadi et al., 2015) and are publicly available at https://github.com/gergely-flamich/relative-entropy-coding. |
| Researcher Affiliation | Collaboration | Gergely Flamich Department of Engineering University of Cambridge gf332@cam.ac.uk Marton Havasi Department of Engineering University of Cambridge mh740@cam.ac.uk José Miguel Hernández-Lobato Department of Engineering University of Cambridge, Microsoft Research, Alan Turing Institute jmh233@cam.ac.uk |
| Pseudocode | Yes | Figure 2: (a) i REC encoder (b) i REC decoder (c) Beam search ensures that log q(z) p(z) is close to the relative entropy. B is the number of beams. (Figure 2 displays structured algorithm blocks for i REC encoder and decoder.) |
| Open Source Code | Yes | Our experiments are implemented in Tensor Flow (Abadi et al., 2015) and are publicly available at https://github.com/gergely-flamich/relative-entropy-coding. |
| Open Datasets | Yes | supported by our empirical results obtained on the Cifar10, Image Net32 and Kodak datasets. Kodak dataset (Eastman Kodak Company, 1999). |
| Dataset Splits | No | We initially set the auxiliary coding distributions by optimizing their variances σ2k on a small validation set to achieve relative entropies close to . For Cifar10 and Image Net32, we used a subsampled test set of size 1000. This indicates usage of validation and test sets but does not provide specific, reproducible split information for training, validation, and test sets or their methodology. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper states 'Our experiments are implemented in Tensor Flow (Abadi et al., 2015)' but does not provide a specific version number for TensorFlow or any other software dependencies. |
| Experiment Setup | Yes | The three hyperparameters of i REC are set to = 3, = 0.2 and B = 20. The hyperparameters of i REC were set this time to = 3, = 0 and B = 10. |