Facial makeup transfer aims to replicate reference makeup on target face, and the existing methods are mainly based on a generic adversarial training process. In this work, we design a Reference-conditional Makeup-aware Discrimination approach (RcMD) to facilitate makeup transfer. Specifically, we perform region-wise semantic feature extraction from a reference makeup image and a source image without makeup. A generator learns to capture and render the reference makeup by modulating the region-wise intermediate features. To ensure precise makeup on target face, we incorporate a reference-conditional discrimination network, which learns to measure the regional makeup consistency between reference and synthesized images. Considering the discrepancy between reference and target faces, an alignment module is trained to fuse the extracted features, conditioned on the reference style. Based on the feature statistics, we perform regional real-synthesized makeup discrimination to ensure precise makeup rendering. Extensive experiments are performed to demonstrate the effectiveness of our designed modules and the superior performance of RcMD in transferring diverse real-world facial makeup.

Although current cross-domain pedestrian detection frame-works have obtained certain positive results, the performance is still source data dependent, which is cumbersome and im-practical in practical applications. To address this issue, we propose a source-free unsupervised pedestrian detection with pseudo label mining and screening. First, a modified CSP de-tector with DropBlock and three detection heads is presented. Then, a multi-expert method is proposed to fuse pseudo la-bels from three detection heads. Finally, a clustering-based self-supervised learning is adopted to categorize pseudo la-bels into positive and negative classes, which forms a set of clusters via similarity of pseudo labels and give classification results based on two confidence scores of each label from the detector backbone and multi-expert fusion. Experimental re-sults on three benchmark datasets show that the proposed approach can achieve state-of-the-art performance and be even comparable with other latest works using source data.

Generative Adversarial Networks (GANs) have shown impressive gains in image synthesis. GAN inversion was recently studied to understand and utilize the knowledge it learns, where a real image is inverted back to a latent code and can thus be reconstructed by the generator. Although increasing the number of latent codes can improve inversion quality to a certain extent, we find that important details may still be neglected when performing feature composition over all the intermediate feature channels. To address this issue, we propose a Prior multi-Subspace Feature Composition (PmSFC) approach for high-fidelity inversion. Considering that the intermediate features are highly correlated with each other, we incorporate a self-expressive layer in the generator to discover meaningful subspaces. In this case, the features at a channel can be expressed as a linear combination of those at other channels in the same subspace. We perform feature composition separately in the subspaces. The semantic differences between them benefit the inversion quality, since the inversion process is regularized based on different aspects of semantics. In the experiments, the superior performance of PmSFC demonstrates the effectiveness of prior subspaces in facilitating GAN inversion together with extended applications in visual manipulation.

One way of addressing the problem of unsupervised domain adaptation (UDA) is to perform adversarial training between two classifiers and their shared feature extractor. The two classifiers are enforced to detect the misaligned regions between the source and target domains, while the feature extractor aligns the features by confusing the classifiers. Although this method yields improvement, it ignores the relationship among target neighbors, which may consequently limit the model performance. In this work, we propose a new alignment strategy based on the "cluster assumption" to ensure the aligned target features preserve their clusters by avoiding overlap with decision boundaries. Furthermore, to make the aligned features more compact, we constrain them to be ro-bust against adversarial perturbation using the different views of the classifiers. Extensive experiments demonstrate the effectiveness of our solution on various datasets.

In this work, we propose an unsupervised ensemble learning method via network generation, referred to as UELNG. Specifically, we first generate weights of clustering ensemble models by adopting HyperGAN, and obtain diverse partitions for data. With these partitions, we can easily identify high-confident pseudo-labels as supervised information to predict low-entropy labels for unlabeled augmented data, thereby enhancing the quality of pseudo-labels and clustering accuracy. We conduct experiments on multiple data sets. Experimental results indicate that our method outperforms state-of-the-art methods by 0.3%, 1.8%, 5.7%, 3.2% and 2.4% on MNIST, STL-10, CIFAR-10, Reuters and 20News, respectively, which demonstrates the effectiveness of our proposed UELNG.

In this paper, we investigate a challenging unsupervised domain adaptation setting --- unsupervised model adaptation. We aim to explore how to rely only on unlabeled target data to improve performance of an existing source prediction model on the target domain, since labeled source data may not be available in some real-world scenarios due to data privacy issues. For this purpose, we propose a new framework, which is referred to as collaborative class conditional generative adversarial net to bypass the dependence on the source data. Specifically, the prediction model is to be improved through generated target-style data, which provides more accurate guidance for the generator. As a result, the generator and the prediction model can collaborate with each other without source data. Furthermore, due to the lack of supervision from source data, we propose a weight constraint that encourages similarity to the source model. A clustering-based regularization is also introduced to produce more discriminative features in the target domain. Compared to conventional domain adaptation methods, our model achieves superior performance on multiple adaptation tasks with only unlabeled target data, which verifies its effectiveness in this challenging setting.