PixelVAE: A Latent Variable Model for Natural Images
Authors: Ishaan Gulrajani, Kundan Kumar, Faruk Ahmed, Adrien Ali Taiga, Francesco Visin, David Vazquez, Aaron Courville
ICLR 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our model achieves state-of-the-art performance on binarized MNIST, competitive performance on 64 64 Image Net, and high-quality samples on the LSUN bedrooms dataset. We evaluate our model on the binarized MNIST dataset (Salakhutdinov & Murray, 2008; Lecun et al., 1998) and report results in Table 1. |
| Researcher Affiliation | Academia | 1 Montreal Institute for Learning Algorithms, Universit e de Montr eal 2 Department of Computer Science and Engineering, IIT Kanpur 3 Centrale Sup elec 4 Computer Vision Center & Universitat Autonoma de Barcelona 5 Politecnico di Milano 6 CIFAR Fellow |
| Pseudocode | No | No explicit pseudocode or algorithm block was found. The model architecture is described in tables in Appendix E. |
| Open Source Code | Yes | We also make an open-source implementation of this model available at https://github.com/igul222/Pixel VAE. |
| Open Datasets | Yes | We evaluate our model on the binarized MNIST dataset (Salakhutdinov & Murray, 2008; Lecun et al., 1998) and We evaluate our model s performance with more data and complicated image distributions, we perform experiments on the LSUN bedrooms dataset (Yu et al., 2015). The 64 64 Image Net generative modeling task was introduced in (van den Oord et al., 2016a). |
| Dataset Splits | No | On 64 64 Image Net, we report validation set likelihood in Table 2. Specific split percentages or counts for validation sets were not provided for any dataset. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, memory, cloud instance types) were provided for the experimental setup, beyond mentioning training "on GPUs". |
| Software Dependencies | No | The authors would like to thank the developers of Theano (Theano Development Team, 2016) and Blocks and Fuel (van Merri enboer et al., 2015). However, explicit version numbers for these software dependencies were not provided. |
| Experiment Setup | Yes | For LSUN and ImageNet models: We optimize using Adam with learning rate 1e-3. Training proceeds for 400K iterations using batch size 48. |