PixelCNN++: Improving the PixelCNN with Discretized Logistic Mixture Likelihood and Other Modifications
Authors: Tim Salimans, Andrej Karpathy, Xi Chen, Diederik P. Kingma
ICLR 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we present state-of-the-art log likelihood results on CIFAR-10 to demonstrate the usefulness of these modifications. |
| Researcher Affiliation | Industry | Tim Salimans, Andrej Karpathy, Xi Chen, Diederik P. Kingma {tim,karpathy,peter,dpkingma}@openai.com |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Here we discuss our implementation of Pixel CNNs which we make available at https://github.com/openai/pixel-cnn. |
| Open Datasets | Yes | We apply our model to modeling natural images in the CIFAR-10 data set. |
| Dataset Splits | No | The paper mentions training on CIFAR-10 and reporting test log-likelihood, but does not explicitly state the specific train/validation/test dataset splits used for reproduction. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions a 'tensorflow implementation' but does not provide specific version numbers for TensorFlow or any other software dependencies. |
| Experiment Setup | Yes | For the encoding part of the Pixel CNN, the model uses 3 Resnet blocks consisting of 5 residual layers, with 2 2 downsampling in between. The same architecture is used for the decoding part of the model, but with upsampling instead of downsampling in between blocks. All residual layers use 192 feature maps and a dropout rate of 0.5. |