Deep Visual Analogy-Making

By Scott E. Reed, Yi Zhang, Yuting Zhang, Honglak Lee. 2015

344 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

A Sober Look at the Unsupervised Learning of Disentangled Representations and their Evaluation (2020, methodology): “Each violin plot shows the distribution of the disentanglement metric across all 50 trained models for each model and hyperparameter setting on Cars3D.”
Generative Adversarial Text to Image Synthesis (2016, methodology): “Reed et al. (2015) encode transformations from analogy pairs, and use a convolutional decoder to predict visual analogies on shapes, video game characters and 3D cars.”
Coupled Generative Adversarial Networks (2016, background): “Our work is also related to the prior works in cross-domain image generation [31, 32, 33], which studied transforming an image in one style to the corresponding images in another style.”
StackGAN: Text to Photo-Realistic Image Synthesis with Stacked Generative Adversarial Networks (2016, background): “Others groups [36, 24] proposed deterministic neural networks as function approximators for image synthesis.”
Autoencoding beyond pixels using a learned similarity metric (2015, methodology): “We refer to our implementation available on-line 1 .”
Representation Learning by Learning to Count (2017, neutral): “[36] exploit an explicit relationship between features.”
On the Transfer of Disentangled Representations in Realistic Settings (2020, background): “Cars3D (Reed et al., 2015) exhibits diﬀerent car models from Fidler et al. (2012) under diﬀerent camera viewpoints.”
Weakly-Supervised Disentanglement Without Compromises (2020, background): “…noise; we assume that noise is explicitly modeled as a latent variable and its effect is manifested through g? as done by (Bengio et al., 2013; Locatello et al., 2019b; Higgins et al., 2018; 2017a; Suter et al., 2019; Reed et al., 2015; LeCun et al., 2004; Kim & Mnih, 2018; Gondal et al., 2019).”
Weakly Supervised Disentanglement with Guarantees (2019, methodology): “We conducted experiments on ﬁve prominent datasets in the disentanglement literature: Shapes3D (Kim & Mnih, 2018), dSprites (Higgins et al., 2017), Scream-dSprites (Locatello et al., 2019), Small-NORB (LeCun et al., 2004), and Cars3D (Reed et al., 2015).”
On the Transfer of Inductive Bias from Simulation to the Real World: a New Disentanglement Dataset (2019, methodology): “For Cars3D,199 CAD models from [11] were used to generate 64x64 color renderings from 24 rotation angles each offset by 15 degrees [41] and available at http://www.”
On the Fairness of Disentangled Representations (2019, background): “Their analysis spans seven datasets: in four of them (dSprites [36], Cars3D [78], SmallNORB [59] and Shapes3D [49]), a deterministic function of the factors of variation is incorporated into the mixing process; they further introduce three additional variants of dSprites, Noisy-dSprites, Color-dSpri”
Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations (2018, methodology): “For this, we plot in Figure 4 (left) the FactorVAE score against different regularization strengths for each model on the Cars3D data set whereas Figure 16 in Appendix I.5 shows the same plot for all data sets and disentanglement metrics.”
Learning Deep Disentangled Embeddings with the F-Statistic Loss (2018, methodology): “For the sprites dataset, we used the encoder architecture of Reed et al. (2015) as well as their embedding dimensionality of 22.”
Disentangling factors of variation in deep representation using adversarial training (2016, methodology): “The problem of disentangling factors of variation also plays an important role in completing image analogies, the goal of the end-to-end model proposed in [24].”
Zero-Shot Task Generalization with Multi-Task Deep Reinforcement Learning (2017, methodology): “Intuitively, the ﬁrst constraint enforces the analogy (i.e., parallelogram structure in the embedding space; see Mikolov et al. (2013); Reed et al. (2015)), while the other constraints prevent trivial solutions.”
Fader Networks: Manipulating Images by Sliding Attributes (2017, background): “This setting was previously explored to learn affine transformations [9], 3D rotations [26], lighting variations [12] and 2D video game animations [20].”
Visual Dynamics: Probabilistic Future Frame Synthesis via Cross Convolutional Networks (2016, methodology): “supervision, our method out-performs the algorithm by Reed et al. [2015], which requires visual analogy labels during training. Visualizing feature maps We visualize the learned feature maps (see Figure 3(b)) in Figure 10. Even without supervision, our network learns to detect objects or contours in”
Disentangled Sequential Autoencoder (2018, methodology): “This dataset comes from an open-source video game project called Liberated Pixel Cup 2 , and has been also considered in Reed et al. (2015); Mathieu et al. (2016) for image processing experiments.”