NeurIPS 2023 论文列表

We present a method for text-driven perpetual view generation -- synthesizing long-term videos of various scenes solely, given an input text prompt describing the scene and camera poses. We introduce a novel framework that generates such videos in an online fashion by combining the generative power of a pre-trained text-to-image model with the geometric priors learned by a pre-trained monocular depth prediction model. To tackle the pivotal challenge of achieving 3D consistency, i.e., synthesizing videos that depict geometrically-plausible scenes, we deploy an online test-time training to encourage the predicted depth map of the current frame to be geometrically consistent with the synthesized scene. The depth maps are used to construct a \emph{unified} mesh representation of the scene, which is progressively constructed along the video generation process. In contrast to previous works, which are applicable only to limited domains, our method generates diverse scenes, such as walkthroughs in spaceships, caves, or ice castles.

Despite the rapid advance of 3D-aware image synthesis, existing studies usually adopt a mixture of techniques and tricks, leaving it unclear how each part contributes to the final performance in terms of generality. Following the most popular and effective paradigm in this field, which incorporates a neural radiance field (NeRF) into the generator of a generative adversarial network (GAN), we build a well-structured codebase through modularizing the generation process. Such a design allows researchers to develop and replace each module independently, and hence offers an opportunity to fairly compare various approaches and recognize their contributions from the module perspective. The reproduction of a range of cutting-edge algorithms demonstrates the availability of our modularized codebase. We also perform a variety of in-depth analyses, such as the comparison across different types of point feature, the necessity of the tailing upsampler in the generator, the reliance on the camera pose prior, etc., which deepen our understanding of existing methods and point out some further directions of the research work. Code and models will be made publicly available to facilitate the development and evaluation of this field.

Real-world data usually confronts severe class-imbalance problems, where several majority classes have a significantly larger presence in the training set than minority classes. One effective solution is using mixup-based methods to generate synthetic samples to enhance the presence of minority classes. Previous approaches mix the background images from the majority classes and foreground images from the minority classes in a random manner, which ignores the sample-level semantic similarity, possibly resulting in less reasonable or less useful images. In this work, we propose an adaptive image-mixing method based on optimal transport (OT) to incorporate both class-level and sample-level information, which is able to generate semantically reasonable and meaningful mixed images for minority classes. Due to its flexibility, our method can be combined with existing long-tailed classification methods to enhance their performance and it can also serve as a general data augmentation method for balanced datasets. Extensive experiments indicate that our method achieves effective performance for long-tailed classification tasks. The code is available at https://github.com/JintongGao/Enhancing-Minority-Classes-by-Mixing.

In an effort to further advance semi-supervised generative and classification tasks, we propose a simple yet effective training strategy called *dual pseudo training* (DPT), built upon strong semi-supervised learners and diffusion models. DPT operates in three stages: training a classifier on partially labeled data to predict pseudo-labels; training a conditional generative model using these pseudo-labels to generate pseudo images; and retraining the classifier with a mix of real and pseudo images. Empirically, DPT consistently achieves SOTA performance of semi-supervised generation and classification across various settings. In particular, with one or two labels per class, DPT achieves a Fréchet Inception Distance (FID) score of 3.08 or 2.52 on ImageNet $256\times256$. Besides, DPT outperforms competitive semi-supervised baselines substantially on ImageNet classification tasks, *achieving top-1 accuracies of 59.0 (+2.8), 69.5 (+3.0), and 74.4 (+2.0)* with one, two, or five labels per class, respectively. Notably, our results demonstrate that diffusion can generate realistic images with only a few labels (e.g., $<0.1$%) and generative augmentation remains viable for semi-supervised classification. Our code is available at *https://github.com/ML-GSAI/DPT*.

Pre-trained large text-to-image models synthesize impressive images with an appropriate use of text prompts. However, ambiguities inherent in natural language, and out-of-distribution effects make it hard to synthesize arbitrary image styles, leveraging a specific design pattern, texture or material. In this paper, we introduce *StyleDrop*, a method that enables the synthesis of images that faithfully follow a specific style using a text-to-image model. StyleDrop is extremely versatile and captures nuances and details of a user-provided style, such as color schemes, shading, design patterns, and local and global effects. StyleDrop works by efficiently learning a new style by fine-tuning very few trainable parameters (less than 1\% of total model parameters), and improving the quality via iterative training with either human or automated feedback. Better yet, StyleDrop is able to deliver impressive results even when the user supplies only a *single* image specifying the desired style. An extensive study shows that, for the task of style tuning text-to-image models, StyleDrop on Muse convincingly outperforms other methods, including DreamBooth and textual inversion on Imagen or Stable Diffusion. More results are available at our project website: [https://styledrop.github.io](https://styledrop.github.io).

Text-to-image diffusion models have recently received a lot of interest for their astonishing ability to produce high-fidelity images from text only. However, achieving one-shot generation that aligns with the user’s intent is nearly impossible, yet small changes to the input prompt often result in very different images. This leaves the user with little semantic control. To put the user in control, we show how to interact with the diffusion process to flexibly steer it along semantic directions. This semantic guidance (SEGA) generalizes to any generative architecture using classifier-free guidance. More importantly, it allows for subtle and extensive edits, composition and style changes, and optimizing the overall artistic conception. We demonstrate SEGA’s effectiveness on both latent and pixel-based diffusion models such as Stable Diffusion, Paella, and DeepFloyd-IF using a variety of tasks, thus providing strong evidence for its versatility and flexibility.

Image restoration involves recovering a high-quality clean image from its degraded version. Deep learning-based methods have significantly improved image restoration performance, however, they have limited generalization ability to different degradation types and levels. This restricts their real-world application since it requires training individual models for each specific degradation and knowing the input degradation type to apply the relevant model. We present a prompt-based learning approach, PromptIR, for All-In-One image restoration that can effectively restore images from various types and levels of degradation. In particular, our method uses prompts to encode degradation-specific information, which is then used to dynamically guide the restoration network. This allows our method to generalize to different degradation types and levels, while still achieving state-of-the-art results on image denoising, deraining, and dehazing. Overall, PromptIR offers a generic and efficient plugin module with few lightweight prompts that can be used to restore images of various types and levels of degradation with no prior information on the corruptions present in the image. Our code and pre-trained models are available here: https://github.com/va1shn9v/PromptIR

Large text-to-image diffusion models have impressive capabilities in generating photorealistic images from text prompts. How to effectively guide or control these powerful models to perform different downstream tasks becomes an important open problem. To tackle this challenge, we introduce a principled finetuning method -- Orthogonal Finetuning (OFT), for adapting text-to-image diffusion models to downstream tasks. Unlike existing methods, OFT can provably preserve hyperspherical energy which characterizes the pairwise neuron relationship on the unit hypersphere. We find that this property is crucial for preserving the semantic generation ability of text-to-image diffusion models. To improve finetuning stability, we further propose Constrained Orthogonal Finetuning (COFT) which imposes an additional radius constraint to the hypersphere. Specifically, we consider two important finetuning text-to-image tasks: subject-driven generation where the goal is to generate subject-specific images given a few images of a subject and a text prompt, and controllable generation where the goal is to enable the model to take in additional control signals. We empirically show that our OFT framework outperforms existing methods in generation quality and convergence speed.

We consider the problem of generating periodic materials with deep models. While symmetry-aware molecule generation has been studied extensively, periodic materials possess different symmetries, which have not been completely captured by existing methods. In this work, we propose SyMat, a novel material generation approach that can capture physical symmetries of periodic material structures. SyMat generates atom types and lattices of materials through generating atom type sets, lattice lengths and lattice angles with a variational auto-encoder model. In addition, SyMat employs a score-based diffusion model to generate atom coordinates of materials, in which a novel symmetry-aware probabilistic model is used in the coordinate diffusion process. We show that SyMat is theoretically invariant to all symmetry transformations on materials and demonstrate that SyMat achieves promising performance on random generation and property optimization tasks. Our code is publicly available as part of the AIRS library (https://github.com/divelab/AIRS).

We present a comprehensive solution to learn and improve text-to-image models from human preference feedback. To begin with, we build ImageReward---the first general-purpose text-to-image human preference reward model---to effectively encode human preferences. Its training is based on our systematic annotation pipeline including rating and ranking, which collects 137k expert comparisons to date. In human evaluation, ImageReward outperforms existing scoring models and metrics, making it a promising automatic metric for evaluating text-to-image synthesis. On top of it, we propose Reward Feedback Learning (ReFL), a direct tuning algorithm to optimize diffusion models against a scorer. Both automatic and human evaluation support ReFL's advantages over compared methods. All code and datasets are provided at \url{https://github.com/THUDM/ImageReward}.

Neural networks are known to be susceptible to adversarial samples: small variations of natural examples crafted to deliberately mislead the models. While they can be easily generated using gradient-based techniques in digital and physical scenarios, they often differ greatly from the actual data distribution of natural images, resulting in a trade-off between strength and stealthiness. In this paper, we propose a novel framework dubbed Diffusion-Based Projected Gradient Descent (Diff-PGD) for generating realistic adversarial samples. By exploiting a gradient guided by a diffusion model, Diff-PGD ensures that adversarial samples remain close to the original data distribution while maintaining their effectiveness. Moreover, our framework can be easily customized for specific tasks such as digital attacks, physical-world attacks, and style-based attacks. Compared with existing methods for generating natural-style adversarial samples, our framework enables the separation of optimizing adversarial loss from other surrogate losses (e.g. content/smoothness/style loss), making it more stable and controllable. Finally, we demonstrate that the samples generated using Diff-PGD have better transferability and anti-purification power than traditional gradient-based methods.

3D segmentation of nuclei images is a fundamental task for many biological studies. Despite the rapid advances of large-volume 3D imaging acquisition methods and the emergence of sophisticated algorithms to segment the nuclei in recent years, a benchmark with all cells completely annotated is still missing, making it hard to accurately assess and further improve the performance of the algorithms. The existing nuclei segmentation benchmarks either worked on 2D only or annotated a small number of 3D cells, perhaps due to the high cost of 3D annotation for large-scale data. To fulfill the critical need, we constructed NIS3D, a 3D, high cell density, large-volume, and completely annotated Nuclei Image Segmentation benchmark, assisted by our newly designed semi-automatic annotation software. NIS3D provides more than 22,000 cells across multiple most-used species in this area. Each cell is labeled by three independent annotators, so we can measure the variability of each annotation. A confidence score is computed for each cell, allowing more nuanced testing and performance comparison. A comprehensive review on the methods of segmenting 3D dense nuclei was conducted. The benchmark was used to evaluate the performance of several selected state-of-the-art segmentation algorithms. The best of current methods is still far away from human-level accuracy, corroborating the necessity of generating such a benchmark. The testing results also demonstrated the strength and weakness of each method and pointed out the directions of further methodological development. The dataset can be downloaded here: https://github.com/yu-lab-vt/NIS3D.

Worldwide Geo-localization aims to pinpoint the precise location of images taken anywhere on Earth. This task has considerable challenges due to the immense variation in geographic landscapes. The image-to-image retrieval-based approaches fail to solve this problem on a global scale as it is not feasible to construct a large gallery of images covering the entire world. Instead, existing approaches divide the globe into discrete geographic cells, transforming the problem into a classification task. However, their performance is limited by the predefined classes and often results in inaccurate localizations when an image's location significantly deviates from its class center. To overcome these limitations, we propose GeoCLIP, a novel CLIP-inspired Image-to-GPS retrieval approach that enforces alignment between the image and its corresponding GPS locations. GeoCLIP's location encoder models the Earth as a continuous function by employing positional encoding through random Fourier features and constructing a hierarchical representation that captures information at varying resolutions to yield a semantically rich high-dimensional feature suitable to use even beyond geo-localization. To the best of our knowledge, this is the first work employing GPS encoding for geo-localization. We demonstrate the efficacy of our method via extensive experiments and ablations on benchmark datasets. We achieve competitive performance with just 20% of training data, highlighting its effectiveness even in limited-data settings. Furthermore, we qualitatively demonstrate geo-localization using a text query by leveraging the CLIP backbone of our image encoder. The project webpage is available at: https://vicentevivan.github.io/GeoCLIP

Blind Omnidirectional Image Quality Assessment (BOIQA) aims to objectively assess the human perceptual quality of omnidirectional images (ODIs) without relying on pristine-quality image information. It is becoming more significant with the increasing advancement of virtual reality (VR) technology. However, the quality assessment of ODIs is severely hampered by the fact that the existing BOIQA pipeline lacks the modeling of the observer's browsing process. To tackle this issue, we propose a novel multi-sequence network for BOIQA called Assessor360, which is derived from the realistic multi-assessor ODI quality assessment procedure. Specifically, we propose a generalized Recursive Probability Sampling (RPS) method for the BOIQA task, combining content and details information to generate multiple pseudo viewport sequences from a given starting point. Additionally, we design a Multi-scale Feature Aggregation (MFA) module with a Distortion-aware Block (DAB) to fuse distorted and semantic features of each viewport. We also devise Temporal Modeling Module (TMM) to learn the viewport transition in the temporal domain. Extensive experimental results demonstrate that Assessor360 outperforms state-of-the-art methods on multiple OIQA datasets. The code and models are available at https://github.com/TianheWu/Assessor360.

Simultaneous sequence generation is a pivotal task for real-time scenarios, such as streaming speech recognition, simultaneous machine translation and simultaneous speech translation, where the target sequence is generated while receiving the source sequence. The crux of achieving high-quality generation with low latency lies in identifying the optimal moments for generating, accomplished by learning a mapping between the source and target sequences. However, existing methods often rely on task-specific heuristics for different sequence types, limiting the model’s capacity to adaptively learn the source-target mapping and hindering the exploration of multi-task learning for various simultaneous tasks. In this paper, we propose a unified segment-to-segment framework (Seg2Seg) for simultaneous sequence generation, which learns the mapping in an adaptive and unified manner. During the process of simultaneous generation, the model alternates between waiting for a source segment and generating a target segment, making the segment serve as the natural bridge between the source and target. To accomplish this, Seg2Seg introduces a latent segment as the pivot between source to target and explores all potential source-target mappings via the proposed expectation training, thereby learning the optimal moments for generating. Experiments on multiple simultaneous generation tasks demonstrate that Seg2Seg achieves state-of-the-art performance and exhibits better generality across various tasks.

Text-driven motion generation has achieved substantial progress with the emergence of diffusion models. However, existing methods still struggle to generate complex motion sequences that correspond to fine-grained descriptions, depicting detailed and accurate spatio-temporal actions.This lack of fine controllability limits the usage of motion generation to a larger audience. To tackle these challenges, we present FineMoGen, a diffusion-based motion generation and editing framework that can synthesize fine-grained motions, with spatial-temporal composition to the user instructions. Specifically, FineMoGen builds upon diffusion model with a novel transformer architecture dubbed Spatio-Temporal Mixture Attention SAMI. SAMI optimizes the generation of the global attention template from two perspectives: 1) explicitly modeling the constraints of spatio-temporal composition; and 2) utilizing sparsely-activated mixture-of-experts to adaptively extract fine-grained features. To facilitate a large-scale study on this new fine-grained motion generation task, we contribute the HuMMan-MoGen dataset, which consists of 2,968 videos and 102,336 fine-grained spatio-temporal descriptions. Extensive experiments validate that FineMoGen exhibits superior motion generation quality over state-of-the-art methods. Notably, FineMoGen further enables zero-shot motion editing capabilities with the aid of modern large language models (LLM), which faithfully manipulates motion sequences with fine-grained instructions.

The zero-shot open-vocabulary setting poses challenges for image classification. Fortunately, utilizing a vision-language model like CLIP, pre-trained on image-text pairs, allows for classifying images by comparing embeddings. Leveraging large language models (LLMs) such as ChatGPT can further enhance CLIP’s accuracy by incorporating class-specific knowledge in descriptions. However, CLIP still exhibits a bias towards certain classes and generates similar descriptions for similar classes, disregarding their differences. To address this problem, we present a novel image classification framework via hierarchical comparisons. By recursively comparing and grouping classes with LLMs, we construct a class hierarchy. With such a hierarchy, we can classify an image by descending from the top to the bottom of the hierarchy, comparing image and text embeddings at each level. Through extensive experiments and analyses, we demonstrate that our proposed approach is intuitive, effective, and explainable. Code will be released upon publication.

We are living in a three-dimensional space while moving forward through a fourth dimension: time. To allow artificial intelligence to develop a comprehensive understanding of such a 4D environment, we introduce **4D Panoptic Scene Graph (PSG-4D)**, a new representation that bridges the raw visual data perceived in a dynamic 4D world and high-level visual understanding. Specifically, PSG-4D abstracts rich 4D sensory data into nodes, which represent entities with precise location and status information, and edges, which capture the temporal relations. To facilitate research in this new area, we build a richly annotated PSG-4D dataset consisting of 3K RGB-D videos with a total of 1M frames, each of which is labeled with 4D panoptic segmentation masks as well as fine-grained, dynamic scene graphs. To solve PSG-4D, we propose PSG4DFormer, a Transformer-based model that can predict panoptic segmentation masks, track masks along the time axis, and generate the corresponding scene graphs via a relation component. Extensive experiments on the new dataset show that our method can serve as a strong baseline for future research on PSG-4D. In the end, we provide a real-world application example to demonstrate how we can achieve dynamic scene understanding by integrating a large language model into our PSG-4D system.

Normalization techniques that capture image style by statistical representation have become a popular component in deep neural networks. Although image enhancement can be considered as a form of style transformation, there has been little exploration of how normalization affect the enhancement performance. To fully leverage the potential of normalization, we present a novel Transition-Constant Normalization (TCN) for various image enhancement tasks. Specifically, it consists of two streams of normalization operations arranged under an invertible constraint, along with a feature sub-sampling operation that satisfies the normalization constraint. TCN enjoys several merits, including being parameter-free, plug-and-play, and incurring no additional computational costs. We provide various formats to utilize TCN for image enhancement, including seamless integration with enhancement networks, incorporation into encoder-decoder architectures for downsampling, and implementation of efficient architectures. Through extensive experiments on multiple image enhancement tasks, like low-light enhancement, exposure correction, SDR2HDR translation, and image dehazing, our TCN consistently demonstrates performance improvements. Besides, it showcases extensive ability in other tasks including pan-sharpening and medical segmentation. The code is available at \textit{\textcolor{blue}{https://github.com/huangkevinj/TCNorm}}.