Ensembling with Deep Generative Views

Chai, Lucy; Zhu, Jun‐Yan; Shechtman, Eli; Isola, Phillip; Zhang, Richard

doi:10.1109/cvpr46437.2021.01475

Cited by 41 publications

(30 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, Conditional Generative Adversarial Networks that generates other similar images or video conditioned on an image, may also be a source of imputation. In particular, [7] was designed for data augmentation.…”

Section: Modeling Hierarchymentioning

confidence: 99%

“…State of the art Conditional GANs do not generalize well to out-of-distribution data. Appendix A.4 contains some examples of generated images from a pretrained network in [7] conditioned on Objectron anchor images. Since extensive experiments are beyond our scope here, we do not investigate GANs further.…”

Section: (B)mentioning

confidence: 99%

“…We attempted to use a pretrained network from [7] to impute scenes from anchor frames in the Objectron test set. The results, shown in Figure 6, show that imputing scenes using a conditional GAN will require fine-tuning, which we defer to future work.…”

Section: A4 Gan Ensembling For Objectronmentioning

confidence: 99%

“…We trained simple feedforward neural networks φ(•) to map the softmax vector, logits, or embedding 7 of an anchor frame x to the scene entropy. Networks that mapped logits and softmax vectors had hidden layers with widths 512, 512, 256, 128, 64 and those that mapped embeddings had hidden layers with widths 4096, 2048, 1024, 512, 256.…”

Section: C3 Auxiliary Networkmentioning

confidence: 99%

“…2 (16) where φ(•) is the auxiliary network, y is its input (logits, softmax vector, or embedding), and σ(•) is the sigmoid function. Results listed in Tables 6 and 7 show that auxiliary networks effectly overfit 7 We refer to the output of the second-to-last year of ResNet-50 and ResNet-101 as the embedding.…”

Section: C3 Auxiliary Networkmentioning

confidence: 99%

See 4 more Smart Citations

Scene Uncertainty and the Wellington Posterior of Deterministic Image Classifiers

Tsuei¹,

Golatkar²,

Soatto³

2021

Preprint

View full text Add to dashboard Cite

We propose a method to estimate the uncertainty of the outcome of an image classifier on a given input datum. Deep neural networks commonly used for image classification are deterministic maps from an input image to an output class. As such, their outcome on a given datum involves no uncertainty, so we must specify what variability we are referring to when defining, measuring and interpreting "confidence." To this end, we introduce the Wellington Posterior, which is the distribution of outcomes that would have been obtained in response to data that could have been generated by the same scene that produced the given image. Since there are infinitely many scenes that could have generated the given image, the Wellington Posterior requires induction from scenes other than the one portrayed. We explore alternate methods using data augmentation, ensembling, and model linearization. Additional alternatives include generative adversarial networks, conditional prior networks, and supervised single-view reconstruction. We test these alternatives against the empirical posterior obtained by inferring the class of temporally adjacent frames in a video. These developments are only a small step towards assessing the reliability of deep network classifiers in a manner that is compatible with safety-critical applications.

show abstract

Section: Modeling Hierarchymentioning

confidence: 99%

Section: (B)mentioning

confidence: 99%

Section: A4 Gan Ensembling For Objectronmentioning

confidence: 99%

Section: C3 Auxiliary Networkmentioning

confidence: 99%

Section: C3 Auxiliary Networkmentioning

confidence: 99%

See 3 more Smart Citations