Variational Interaction Information Maximization for Cross-domain Disentanglement

Authors: HyeongJoo Hwang, Geon-Hyeong Kim, Seunghoon Hong, Kee-Eung Kim

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the validity of our model in the image-to-image translation and the cross-domain retrieval tasks. We further show that our model achieves the state-of-the-art performance in the zero-shot sketch based image retrieval task, even without external knowledge. 4 Experiments We employ experiments on image-to-image translation and image retrieval tasks to evaluate the quality of cross-domain disentanglement.
Researcher Affiliation Academia Hyeong Joo Hwang1, Geon-Hyeong Kim2, Seunghoon Hong2, Kee-Eung Kim1,2 1 Graduate School of AI, KAIST, Daejeon, Republic of Korea 2 School of Computing, KAIST, Daejeon, Republic of Korea
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No We leave all the implementation details and hyperparameter settings in D in the supplementary material.
Open Datasets Yes Datasets We evaluate our method on two datasets: MNIST-CDCB [12] and Cars [36] datasets. ... We tested our model with MNIST-CDCB [12], Facades [41], and Maps [17] datasets. ... We evaluate our model on Sketchy (Extended) [29, 37], one of the most widely used datasets of sketch and photo images in sketch-based image retrieval (SBIR) task.
Dataset Splits Yes In MNIST-CDCB [12] dataset... We use 50,000 / 10,000 pairs of train/test samples following [24]. Cars [36]... Out of those 16,836 pairs of 183 cars, we assigned 16,192 pairs of 176 cars to train set and 644 pairs of 7 cars to test set. ... In Facades [41] dataset... We use 400 / 100 / 106 pairs of train/valid/test samples following [41]. In Maps [17] dataset... We use 1096 / 1098 pairs of train/test samples following [17].
Hardware Specification No The paper does not provide specific hardware details for running experiments.
Software Dependencies No extracted features of images from VGG16 and finetuned with the train set of Sketchy Extended.
Experiment Setup No We leave all the implementation details and hyperparameter settings in D in the supplementary material.