Jie An

Siyu Huang, Jie An, Donglai Wei et al. · harvard

Unpaired image-to-image translation (UNIT) aims to map images between two visual domains without paired training data. However, given a UNIT model trained on certain domains, it is difficult for current methods to incorporate new domains because they often need to train the full model on both existing and new domains. To address this problem, we propose a new domain-scalable UNIT method, termed as latent space anchoring, which can be efficiently extended to new visual domains and does not need to fine-tune encoders and decoders of existing domains. Our method anchors images of different domains to the same latent space of frozen GANs by learning lightweight encoder and regressor models to reconstruct single-domain images. In the inference phase, the learned encoders and decoders of different domains can be arbitrarily combined to translate images between any two domains without fine-tuning. Experiments on various datasets show that the proposed method achieves superior performance on both standard and domain-scalable UNIT tasks in comparison with the state-of-the-art methods.

Alexander Martin, Haitian Zheng, Jie An et al.

With a strong understanding of the target domain from natural language, we produce promising results in translating across large domain gaps and bringing skeletons back to life. In this work, we use text-guided latent diffusion models for zero-shot image-to-image translation (I2I) across large domain gaps (longI2I), where large amounts of new visual features and new geometry need to be generated to enter the target domain. Being able to perform translations across large domain gaps has a wide variety of real-world applications in criminology, astrology, environmental conservation, and paleontology. In this work, we introduce a new task Skull2Animal for translating between skulls and living animals. On this task, we find that unguided Generative Adversarial Networks (GANs) are not capable of translating across large domain gaps. Instead of these traditional I2I methods, we explore the use of guided diffusion and image editing models and provide a new benchmark model, Revive-2I, capable of performing zero-shot I2I via text-prompting latent diffusion models. We find that guidance is necessary for longI2I because, to bridge the large domain gap, prior knowledge about the target domain is needed. In addition, we find that prompting provides the best and most scalable information about the target domain as classifier-guided diffusion models require retraining for specific use cases and lack stronger constraints on the target domain because of the wide variety of images they are trained on.

Jingyuan Chen, Fuchen Long, Jie An et al.

The first-in-first-out (FIFO) video diffusion, built on a pre-trained text-to-video model, has recently emerged as an effective approach for tuning-free long video generation. This technique maintains a queue of video frames with progressively increasing noise, continuously producing clean frames at the queue's head while Gaussian noise is enqueued at the tail. However, FIFO-Diffusion often struggles to keep long-range temporal consistency in the generated videos due to the lack of correspondence modeling across frames. In this paper, we propose Ouroboros-Diffusion, a novel video denoising framework designed to enhance structural and content (subject) consistency, enabling the generation of consistent videos of arbitrary length. Specifically, we introduce a new latent sampling technique at the queue tail to improve structural consistency, ensuring perceptually smooth transitions among frames. To enhance subject consistency, we devise a Subject-Aware Cross-Frame Attention (SACFA) mechanism, which aligns subjects across frames within short segments to achieve better visual coherence. Furthermore, we introduce self-recurrent guidance. This technique leverages information from all previous cleaner frames at the front of the queue to guide the denoising of noisier frames at the end, fostering rich and contextual global information interaction. Extensive experiments of long video generation on the VBench benchmark demonstrate the superiority of our Ouroboros-Diffusion, particularly in terms of subject consistency, motion smoothness, and temporal consistency.

Zhenghong Zhou, Jie An, Jiebo Luo

Precise camera pose control is crucial for video generation with diffusion models. Existing methods require fine-tuning with additional datasets containing paired videos and camera pose annotations, which are both data-intensive and computationally costly, and can disrupt the pre-trained model distribution. We introduce Latent-Reframe, which enables camera control in a pre-trained video diffusion model without fine-tuning. Unlike existing methods, Latent-Reframe operates during the sampling stage, maintaining efficiency while preserving the original model distribution. Our approach reframes the latent code of video frames to align with the input camera trajectory through time-aware point clouds. Latent code inpainting and harmonization then refine the model latent space, ensuring high-quality video generation. Experimental results demonstrate that Latent-Reframe achieves comparable or superior camera control precision and video quality to training-based methods, without the need for fine-tuning on additional datasets.

Jie An, De Wang, Pengsheng Guo et al.

Recent work studying the generalization of diffusion models with UNet-based denoisers reveals inductive biases that can be expressed via geometry-adaptive harmonic bases. However, in practice, more recent denoising networks are often based on transformers, e.g., the diffusion transformer (DiT). This raises the question: do transformer-based denoising networks exhibit inductive biases that can also be expressed via geometry-adaptive harmonic bases? To our surprise, we find that this is not the case. This discrepancy motivates our search for the inductive bias that can lead to good generalization in DiT models. Investigating the pivotal attention modules of a DiT, we find that locality of attention maps are closely associated with generalization. To verify this finding, we modify the generalization of a DiT by restricting its attention windows. We inject local attention windows to a DiT and observe an improvement in generalization. Furthermore, we empirically find that both the placement and the effective attention size of these local attention windows are crucial factors. Experimental results on the CelebA, ImageNet, and LSUN datasets show that strengthening the inductive bias of a DiT can improve both generalization and generation quality when less training data is available. Source code will be released publicly upon paper publication. Project page: dit-generalization.github.io/.

Xiaochen Tang, Zhenya Zhang, Miaomiao Zhang et al.

In real-time and safety-critical cyber-physical systems (CPSs), control synthesis must guarantee that generated policies meet stringent timing and correctness requirements under uncertain and dynamic conditions. Signal temporal logic (STL) has emerged as a powerful formalism of expressing real-time constraints, with its semantics enabling quantitative assessment of system behavior. Meanwhile, reinforcement learning (RL) has become an important method for solving control synthesis problems in unknown environments. Recent studies incorporate STL-based reward functions into RL to automatically synthesize control policies. However, the automatically inferred rewards obtained by these methods represent the global assessment of a whole or partial path but do not accumulate the rewards of local changes accurately, so the sparse global rewards may lead to non-convergence and unstable training performances. In this paper, we propose an online reward generation method guided by the online causation monitoring of STL. Our approach continuously monitors system behavior against an STL specification at each control step, computing the quantitative distance toward satisfaction or violation and thereby producing rewards that reflect instantaneous state dynamics. Additionally, we provide a smooth approximation of the causation semantics to overcome the discontinuity of the causation semantics and make it differentiable for using deep-RL methods. We have implemented a prototype tool and evaluated it in the Gym environment on a variety of continuously controlled benchmarks. Experimental results show that our proposed STL-guided RL method with online causation semantics outperforms existing relevant STL-guided RL methods, providing a more robust and efficient reward generation framework for deep-RL.

Jie An, Siyu Huang, Yibing Song et al.

Universal style transfer retains styles from reference images in content images. While existing methods have achieved state-of-the-art style transfer performance, they are not aware of the content leak phenomenon that the image content may corrupt after several rounds of stylization process. In this paper, we propose ArtFlow to prevent content leak during universal style transfer. ArtFlow consists of reversible neural flows and an unbiased feature transfer module. It supports both forward and backward inferences and operates in a projection-transfer-reversion scheme. The forward inference projects input images into deep features, while the backward inference remaps deep features back to input images in a lossless and unbiased way. Extensive experiments demonstrate that ArtFlow achieves comparable performance to state-of-the-art style transfer methods while avoiding content leak.

Jie An, Tianlang Chen, Songyang Zhang et al.

Transferring the sentiment of an image is an unexplored research topic in the area of computer vision. This work proposes a novel framework consisting of a reference image retrieval step and a global sentiment transfer step to transfer sentiments of images according to a given sentiment tag. The proposed image retrieval algorithm is based on the SSIM index. The retrieved reference images by the proposed algorithm are more content-related against the algorithm based on the perceptual loss. Therefore can lead to a better image sentiment transfer result. In addition, we propose a global sentiment transfer step, which employs an optimization algorithm to iteratively transfer sentiment of images based on feature maps produced by the Densenet121 architecture. The proposed sentiment transfer algorithm can transfer the sentiment of images while ensuring the content structure of the input image intact. The qualitative and quantitative experiments demonstrate that the proposed sentiment transfer framework outperforms existing artistic and photorealistic style transfer algorithms in making reliable sentiment transfer results with rich and exact details.

Jie An, Tao Li, Haozhi Huang et al.

Extracting effective deep features to represent content and style information is the key to universal style transfer. Most existing algorithms use VGG19 as the feature extractor, which incurs a high computational cost and impedes real-time style transfer on high-resolution images. In this work, we propose a lightweight alternative architecture - ArtNet, which is based on GoogLeNet, and later pruned by a novel channel pruning method named Zero-channel Pruning specially designed for style transfer approaches. Besides, we propose a theoretically sound sandwich swap transform (S2) module to transfer deep features, which can create a pleasing holistic appearance and good local textures with an improved content preservation ability. By using ArtNet and S2, our method is 2.3 to 107.4 times faster than state-of-the-art approaches. The comprehensive experiments demonstrate that ArtNet can achieve universal, real-time, and high-quality style transfer on high-resolution images simultaneously, (68.03 FPS on 512 times 512 images).

Jie An, Haoyi Xiong, Jun Huan et al.

The key challenge in photorealistic style transfer is that an algorithm should faithfully transfer the style of a reference photo to a content photo while the generated image should look like one captured by a camera. Although several photorealistic style transfer algorithms have been proposed, they need to rely on post- and/or pre-processing to make the generated images look photorealistic. If we disable the additional processing, these algorithms would fail to produce plausible photorealistic stylization in terms of detail preservation and photorealism. In this work, we propose an effective solution to these issues. Our method consists of a construction step (C-step) to build a photorealistic stylization network and a pruning step (P-step) for acceleration. In the C-step, we propose a dense auto-encoder named PhotoNet based on a carefully designed pre-analysis. PhotoNet integrates a feature aggregation module (BFA) and instance normalized skip links (INSL). To generate faithful stylization, we introduce multiple style transfer modules in the decoder and INSLs. PhotoNet significantly outperforms existing algorithms in terms of both efficiency and effectiveness. In the P-step, we adopt a neural architecture search method to accelerate PhotoNet. We propose an automatic network pruning framework in the manner of teacher-student learning for photorealistic stylization. The network architecture named PhotoNAS resulted from the search achieves significant acceleration over PhotoNet while keeping the stylization effects almost intact. We conduct extensive experiments on both image and video transfer. The results show that our method can produce favorable results while achieving 20-30 times acceleration in comparison with the existing state-of-the-art approaches. It is worth noting that the proposed algorithm accomplishes better performance without any pre- or post-processing.

Jie An, Haoyi Xiong, Jiebo Luo et al.

Universal style transfer is an image editing task that renders an input content image using the visual style of arbitrary reference images, including both artistic and photorealistic stylization. Given a pair of images as the source of content and the reference of style, existing solutions usually first train an auto-encoder (AE) to reconstruct the image using deep features and then embeds pre-defined style transfer modules into the AE reconstruction procedure to transfer the style of the reconstructed image through modifying the deep features. While existing methods typically need multiple rounds of time-consuming AE reconstruction for better stylization, our work intends to design novel neural network architectures on top of AE for fast style transfer with fewer artifacts and distortions all in one pass of end-to-end inference. To this end, we propose two network architectures named ArtNet and PhotoNet to improve artistic and photo-realistic stylization, respectively. Extensive experiments demonstrate that ArtNet generates images with fewer artifacts and distortions against the state-of-the-art artistic transfer algorithms, while PhotoNet improves the photorealistic stylization results by creating sharp images faithfully preserving rich details of the input content. Moreover, ArtNet and PhotoNet can achieve 3X to 100X speed-up over the state-of-the-art algorithms, which is a major advantage for large content images.

Jie An, Haoyi Xiong, Jinwen Ma et al.

Neural Architecture Search (NAS) has been widely studied for designing discriminative deep learning models such as image classification, object detection, and semantic segmentation. As a large number of priors have been obtained through the manual design of architectures in the fields, NAS is usually considered as a supplement approach. In this paper, we have significantly expanded the application areas of NAS by performing an empirical study of NAS to search generative models, or specifically, auto-encoder based universal style transfer, which lacks systematic exploration, if any, from the architecture search aspect. In our work, we first designed a search space where common operators for image style transfer such as VGG-based encoders, whitening and coloring transforms (WCT), convolution kernels, instance normalization operators, and skip connections were searched in a combinatorial approach. With a simple yet effective parallel evolutionary NAS algorithm with multiple objectives, we derived the first group of end-to-end deep networks for universal photorealistic style transfer. Comparing to random search, a NAS method that is gaining popularity recently, we demonstrated that carefully designed search strategy leads to much better architecture design. Finally compared to existing universal style transfer networks for photorealistic rendering such as PhotoWCT that stacks multiple well-trained auto-encoders and WCT transforms in a non-end-to-end manner, the architectures designed by StyleNAS produce better style-transferred images with details preserving, using a tiny number of operators/parameters, and enjoying around 500x inference time speed-up.