Randomized Autoregressive Visual Generation

This paper presents Randomized AutoRegressive modeling (RAR) for visual generation, which sets a new state-of-the-art performance on the image generation task while maintaining full compatibility with language modeling frameworks. The proposed RAR is simple: during a standard autoregressive training process with a next-token prediction objective, the input sequence-typically ordered in raster form-is randomly permuted into different factorization orders with a probability r, where r starts at 1 and linearly decays to 0 over the course of training. This annealing training strategy enables the model to learn to maximize the expected likelihood over all factorization orders and thus effectively improve the model's capability of modeling bidirectional contexts. Importantly, RAR preserves the integrity of the autoregressive modeling framework, ensuring full compatibility with language modeling while significantly improving performance in image generation. On the ImageNet-256 benchmark, RAR achieves an FID score of 1.48, not only surpassing prior state-of-the-art autoregressive image generators but also outperforming leading diffusion-based and masked transformer-based methods. Code and models will be made available at https://github.com/bytedance/1d-tokenizer

LLaVolta: Efficient Multi-modal Models via Stage-wise Visual Context Compression

While significant advancements have been made in compressed representations for text embeddings in large language models (LLMs), the compression of visual tokens in large multi-modal models (LMMs) has remained a largely overlooked area. In this work, we present the study on the analysis of redundancy concerning visual tokens and efficient training within these models. Our initial experiments show that eliminating up to 70% of visual tokens at the testing stage by simply average pooling only leads to a minimal 3% reduction in visual question answering accuracy on the GQA benchmark, indicating significant redundancy in visual context. Addressing this, we introduce Visual Context Compressor, which reduces the number of visual tokens during training to enhance training efficiency without sacrificing performance. To minimize information loss caused by the compression on visual tokens while maintaining training efficiency, we develop LLaVolta as a lite training scheme. LLaVolta incorporates stage-wise visual context compression to progressively compress the visual tokens from heavily to lightly, and finally no compression at the end of training, yielding no loss of information when testing. Extensive experiments demonstrate that our approach enhances the performance of MLLMs in both image-language and video-language understanding, while also significantly cutting training costs. Code is available at https://github.com/Beckschen/LLaVolta

Alleviating Distortion in Image Generation via Multi-Resolution Diffusion Models

This paper presents innovative enhancements to diffusion models by integrating a novel multi-resolution network and time-dependent layer normalization. Diffusion models have gained prominence for their effectiveness in high-fidelity image generation. While conventional approaches rely on convolutional U-Net architectures, recent Transformer-based designs have demonstrated superior performance and scalability. However, Transformer architectures, which tokenize input data (via "patchification"), face a trade-off between visual fidelity and computational complexity due to the quadratic nature of self-attention operations concerning token length. While larger patch sizes enable attention computation efficiency, they struggle to capture fine-grained visual details, leading to image distortions. To address this challenge, we propose augmenting the Diffusion model with the Multi-Resolution network (DiMR), a framework that refines features across multiple resolutions, progressively enhancing detail from low to high resolution. Additionally, we introduce Time-Dependent Layer Normalization (TD-LN), a parameter-efficient approach that incorporates time-dependent parameters into layer normalization to inject time information and achieve superior performance. Our method's efficacy is demonstrated on the class-conditional ImageNet generation benchmark, where DiMR-XL variants outperform prior diffusion models, setting new state-of-the-art FID scores of 1.70 on ImageNet 256 x 256 and 2.89 on ImageNet 512 x 512. Project page: https://qihao067.github.io/projects/DiMR

MaXTron: Mask Transformer with Trajectory Attention for Video Panoptic Segmentation

Video panoptic segmentation requires consistently segmenting (for both `thing' and `stuff' classes) and tracking objects in a video over time. In this work, we present MaXTron, a general framework that exploits Mask XFormer with Trajectory Attention to tackle the task. MaXTron enriches an off-the-shelf mask transformer by leveraging trajectory attention. The deployed mask transformer takes as input a short clip consisting of only a few frames and predicts the clip-level segmentation. To enhance the temporal consistency, MaXTron employs within-clip and cross-clip tracking modules, efficiently utilizing trajectory attention. Originally designed for video classification, trajectory attention learns to model the temporal correspondences between neighboring frames and aggregates information along the estimated motion paths. However, it is nontrivial to directly extend trajectory attention to the per-pixel dense prediction tasks due to its quadratic dependency on input size. To alleviate the issue, we propose to adapt the trajectory attention for both the dense pixel features and object queries, aiming to improve the short-term and long-term tracking results, respectively. Particularly, in our within-clip tracking module, we propose axial-trajectory attention that effectively computes the trajectory attention for tracking dense pixels sequentially along the height- and width-axes. The axial decomposition significantly reduces the computational complexity for dense pixel features. In our cross-clip tracking module, since the object queries in mask transformer are learned to encode the object information, we are able to capture the long-term temporal connections by applying trajectory attention to object queries, which learns to track each object across different clips. Without bells and whistles, MaXTron demonstrates state-of-the-art performances on video segmentation benchmarks.

Learning Part Segmentation from Synthetic Animals

Semantic part segmentation provides an intricate and interpretable understanding of an object, thereby benefiting numerous downstream tasks. However, the need for exhaustive annotations impedes its usage across diverse object types. This paper focuses on learning part segmentation from synthetic animals, leveraging the Skinned Multi-Animal Linear (SMAL) models to scale up existing synthetic data generated by computer-aided design (CAD) animal models. Compared to CAD models, SMAL models generate data with a wider range of poses observed in real-world scenarios. As a result, our first contribution is to construct a synthetic animal dataset of tigers and horses with more pose diversity, termed Synthetic Animal Parts (SAP). We then benchmark Syn-to-Real animal part segmentation from SAP to PartImageNet, namely SynRealPart, with existing semantic segmentation domain adaptation methods and further improve them as our second contribution. Concretely, we examine three Syn-to-Real adaptation methods but observe relative performance drop due to the innate difference between the two tasks. To address this, we propose a simple yet effective method called Class-Balanced Fourier Data Mixing (CB-FDM). Fourier Data Mixing aligns the spectral amplitudes of synthetic images with real images, thereby making the mixed images have more similar frequency content to real images. We further use Class-Balanced Pseudo-Label Re-Weighting to alleviate the imbalanced class distribution. We demonstrate the efficacy of CB-FDM on SynRealPart over previous methods with significant performance improvements. Remarkably, our third contribution is to reveal that the learned parts from synthetic tiger and horse are transferable across all quadrupeds in PartImageNet, further underscoring the utility and potential applications of animal part segmentation.

Convolutions Die Hard: Open-Vocabulary Segmentation with Single Frozen Convolutional CLIP

Open-vocabulary segmentation is a challenging task requiring segmenting and recognizing objects from an open set of categories. One way to address this challenge is to leverage multi-modal models, such as CLIP, to provide image and text features in a shared embedding space, which bridges the gap between closed-vocabulary and open-vocabulary recognition. Hence, existing methods often adopt a two-stage framework to tackle the problem, where the inputs first go through a mask generator and then through the CLIP model along with the predicted masks. This process involves extracting features from images multiple times, which can be ineffective and inefficient. By contrast, we propose to build everything into a single-stage framework using a shared Frozen Convolutional CLIP backbone, which not only significantly simplifies the current two-stage pipeline, but also remarkably yields a better accuracy-cost trade-off. The proposed FC-CLIP, benefits from the following observations: the frozen CLIP backbone maintains the ability of open-vocabulary classification and can also serve as a strong mask generator, and the convolutional CLIP generalizes well to a larger input resolution than the one used during contrastive image-text pretraining. When training on COCO panoptic data only and testing in a zero-shot manner, FC-CLIP achieve 26.8 PQ, 16.8 AP, and 34.1 mIoU on ADE20K, 18.2 PQ, 27.9 mIoU on Mapillary Vistas, 44.0 PQ, 26.8 AP, 56.2 mIoU on Cityscapes, outperforming the prior art by +4.2 PQ, +2.4 AP, +4.2 mIoU on ADE20K, +4.0 PQ on Mapillary Vistas and +20.1 PQ on Cityscapes, respectively. Additionally, the training and testing time of FC-CLIP is 7.5x and 6.6x significantly faster than the same prior art, while using 5.9x fewer parameters. FC-CLIP also sets a new state-of-the-art performance across various open-vocabulary semantic segmentation datasets. Code at https://github.com/bytedance/fc-clip

Compositor: Bottom-up Clustering and Compositing for Robust Part and Object Segmentation

In this work, we present a robust approach for joint part and object segmentation. Specifically, we reformulate object and part segmentation as an optimization problem and build a hierarchical feature representation including pixel, part, and object-level embeddings to solve it in a bottom-up clustering manner. Pixels are grouped into several clusters where the part-level embeddings serve as cluster centers. Afterwards, object masks are obtained by compositing the part proposals. This bottom-up interaction is shown to be effective in integrating information from lower semantic levels to higher semantic levels. Based on that, our novel approach Compositor produces part and object segmentation masks simultaneously while improving the mask quality. Compositor achieves state-of-the-art performance on PartImageNet and Pascal-Part by outperforming previous methods by around 0.9% and 1.3% on PartImageNet, 0.4% and 1.7% on Pascal-Part in terms of part and object mIoU and demonstrates better robustness against occlusion by around 4.4% and 7.1% on part and object respectively. Code will be available at https://github.com/TACJu/Compositor.

Learning from Temporal Gradient for Semi-supervised Action Recognition

Semi-supervised video action recognition tends to enable deep neural networks to achieve remarkable performance even with very limited labeled data. However, existing methods are mainly transferred from current image-based methods (e.g., FixMatch). Without specifically utilizing the temporal dynamics and inherent multimodal attributes, their results could be suboptimal. To better leverage the encoded temporal information in videos, we introduce temporal gradient as an additional modality for more attentive feature extraction in this paper. To be specific, our method explicitly distills the fine-grained motion representations from temporal gradient (TG) and imposes consistency across different modalities (i.e., RGB and TG). The performance of semi-supervised action recognition is significantly improved without additional computation or parameters during inference. Our method achieves the state-of-the-art performance on three video action recognition benchmarks (i.e., Kinetics-400, UCF-101, and HMDB-51) under several typical semi-supervised settings (i.e., different ratios of labeled data).

ROBIN : A Benchmark for Robustness to Individual Nuisances in Real-World Out-of-Distribution Shifts

Enhancing the robustness in real-world scenarios has been proven very challenging. One reason is that existing robustness benchmarks are limited, as they either rely on synthetic data or they simply measure robustness as generalization between datasets and hence ignore the effects of individual nuisance factors. In this work, we introduce ROBIN, a benchmark dataset for diagnosing the robustness of vision algorithms to individual nuisances in real-world images. ROBIN builds on 10 rigid categories from the PASCAL VOC 2012 and ImageNet datasets and includes out-of-distribution examples of the objects 3D pose, shape, texture, context and weather conditions. ROBIN is richly annotated to enable benchmark models for image classification, object detection, and 3D pose estimation. We provide results for a number of popular baselines and make several interesting observations: 1. Some nuisance factors have a much stronger negative effect on the performance compared to others. Moreover, the negative effect of an OODnuisance depends on the downstream vision task. 2. Current approaches to enhance OOD robustness using strong data augmentation have only marginal effects in real-world OOD scenarios, and sometimes even reduce the OOD performance. 3. We do not observe any significant differences between convolutional and transformer architectures in terms of OOD robustness. We believe our dataset provides a rich testbed to study the OOD robustness of vision algorithms and will help to significantly push forward research in this area.

PartImageNet: A Large, High-Quality Dataset of Parts

A part-based object understanding facilitates efficient compositional learning and knowledge transfer, robustness to occlusion, and has the potential to increase the performance on general recognition and localization tasks. However, research on part-based models is hindered due to the lack of datasets with part annotations, which is caused by the extreme difficulty and high cost of annotating object parts in images. In this paper, we propose PartImageNet, a large, high-quality dataset with part segmentation annotations. It consists of 158 classes from ImageNet with approximately 24000 images. PartImageNet is unique because it offers part-level annotations on a general set of classes with non-rigid, articulated objects, while having an order of magnitude larger size compared to existing datasets. It can be utilized in multiple vision tasks including but not limited to: Part Discovery, Semantic Segmentation, Few-shot Learning. Comprehensive experiments are conducted to set up a set of baselines on PartImageNet and we find that existing works on part discovery can not always produce satisfactory results during complex variations. The exploit of parts on downstream tasks also remains insufficient. We believe that our PartImageNet will greatly facilitate the research on part-based models and their applications. The dataset and scripts will soon be released at https://github.com/TACJu/PartImageNet.