Adaptive IMLE for Few-shot Pretraining-free Generative Modelling

Authors: Mehran Aghabozorgi, Shichong Peng, Ke Li

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate on multiple few-shot image synthesis datasets that our method significantly outperforms existing methods. Our code is available at https: //github.com/mehranagh20/Ada IMLE.
Researcher Affiliation Academia 1APEX Lab, School of Computing Science, Simon Fraser University. Correspondence to: Mehran Aghabozorgi <mehran_aghabozorgi@sfu.ca>, Shichong Peng <shichong_peng@sfu.ca>, Ke Li <keli@sfu.ca>.
Pseudocode Yes Algorithm 1 Vanilla IMLE procedure
Open Source Code Yes Our code is available at https: //github.com/mehranagh20/Ada IMLE.
Open Datasets Yes We evaluate our method and the baselines on a wide range of natural image datasets at 256 256 resolution, which includes Animal-Face Dog and Cat (Si & Zhu, 2012), Obama, Panda, and Grumpy-cat (Zhao et al., 2020b) and Flickr-Face HQ (FFHQ) subset (Karras et al., 2019).
Dataset Splits No For LPIPS backtracking, we use 90% of the full dataset for training and evaluate the metric using the remaining 10% of the dataset. This describes a train/test split, but no separate validation split for hyperparameter tuning is explicitly mentioned.
Hardware Specification Yes We train our model for less than 200k iterations with a mini-batch size of 4 using the Adam optimizer (Kingma & Ba, 2015) with a learning rate of 2 10 6 on a single NVIDIA V100 GPU.
Software Dependencies No The paper mentions using Adam optimizer but does not specify versions of software libraries (e.g., PyTorch, TensorFlow, Python) or other specific ancillary software with version numbers.
Experiment Setup Yes Our network architecture is modified from (Child, 2021), where we keep the decoder architecture and replace the encoder with a fully-connected mapping network inspired by (Karras et al., 2019). We choose an input latent dimension of 1024, m = 10000 and a tightening coefficient δ = 0.98. We train our model for less than 200k iterations with a mini-batch size of 4 using the Adam optimizer (Kingma & Ba, 2015) with a learning rate of 2 10 6 on a single NVIDIA V100 GPU.