We present an approach to pose object recognition as next token prediction. The idea is to apply a language decoder that auto-regressively predicts the text tokens from image embeddings to form labels. To ground this prediction process in auto-regression, we customize a non-causal attention mask for the decoder, incorporating two key features: modeling tokens from different labels to be independent, and treating image tokens as a prefix. This masking mechanism inspires an efficient method - one-shot sampling - to simultaneously sample tokens of multiple labels in parallel and rank generated labels by their probabilities during inference. To further enhance the efficiency, we propose a simple strategy to construct a compact decoder by simply discarding the intermediate blocks of a pretrained language model. This approach yields a decoder that matches the full model's performance while being notably more efficient. The code is available at https://github.com/kaiyuyue/nxtp
We introduce Generative Infinite-Vocabulary Transformers (GIVT) which generate vector sequences with real-valued entries, instead of discrete tokens from a finite vocabulary. To this end, we propose two surprisingly simple modifications to decoder-only transformers: 1) at the input, we replace the finite-vocabulary lookup table with a linear projection of the input vectors; and 2) at the output, we replace the logits prediction (usually mapped to a categorical distribution) with the parameters of a multivariate Gaussian mixture model. Inspired by the image-generation paradigm of VQ-GAN and MaskGIT, where transformers are used to model the discrete latent sequences of a VQ-VAE, we use GIVT to model the unquantized real-valued latent sequences of a $\beta$-VAE. In class-conditional image generation GIVT outperforms VQ-GAN (and improved variants thereof) as well as MaskGIT, and achieves performance competitive with recent latent diffusion models. Finally, we obtain strong results outside of image generation when applying GIVT to panoptic segmentation and depth estimation with a VAE variant of the UViM framework.
Image super-resolution aims to synthesize high-resolution image from a low-resolution image. It is an active area to overcome the resolution limitations in several applications like low-resolution object-recognition, medical image enhancement, etc. The generative adversarial network (GAN) based methods have been the state-of-the-art for image super-resolution by utilizing the convolutional neural networks (CNNs) based generator and discriminator networks. However, the CNNs are not able to exploit the global information very effectively in contrast to the transformers, which are the recent breakthrough in deep learning by exploiting the self-attention mechanism. Motivated from the success of transformers in language and vision applications, we propose a SRTransGAN for image super-resolution using transformer based GAN. Specifically, we propose a novel transformer-based encoder-decoder network as a generator to generate 2x images and 4x images. We design the discriminator network using vision transformer which uses the image as sequence of patches and hence useful for binary classification between synthesized and real high-resolution images. The proposed SRTransGAN outperforms the existing methods by 4.38 % on an average of PSNR and SSIM scores. We also analyze the saliency map to understand the learning ability of the proposed method.
The autonomous driving community has shown significant interest in 3D occupancy prediction, driven by its exceptional geometric perception and general object recognition capabilities. To achieve this, current works try to construct a Tri-Perspective View (TPV) or Occupancy (OCC) representation extending from the Bird-Eye-View perception. However, compressed views like TPV representation lose 3D geometry information while raw and sparse OCC representation requires heavy but redundant computational costs. To address the above limitations, we propose Compact Occupancy TRansformer (COTR), with a geometry-aware occupancy encoder and a semantic-aware group decoder to reconstruct a compact 3D OCC representation. The occupancy encoder first generates a compact geometrical OCC feature through efficient explicit-implicit view transformation. Then, the occupancy decoder further enhances the semantic discriminability of the compact OCC representation by a coarse-to-fine semantic grouping strategy. Empirical experiments show that there are evident performance gains across multiple baselines, e.g., COTR outperforms baselines with a relative improvement of 8%-15%, demonstrating the superiority of our method.
Human-Object Interaction (HOI) detection is a core task for human-centric image understanding. Recent one-stage methods adopt a transformer decoder to collect image-wide cues that are useful for interaction prediction; however, the interaction representations obtained using this method are entangled and lack interpretability. In contrast, traditional two-stage methods benefit significantly from their ability to compose interaction features in a disentangled and explainable manner. In this paper, we improve the performance of one-stage methods by enabling them to extract disentangled interaction representations. First, we propose Shunted Cross-Attention (SCA) to extract human appearance, object appearance, and global context features using different cross-attention heads. This is achieved by imposing different masks on the cross-attention maps produced by the different heads. Second, we introduce the Interaction-aware Pose Estimation (IPE) task to learn interaction-relevant human pose features using a disentangled decoder. This is achieved with a novel attention module that accurately captures the human keypoints relevant to the current interaction category. Finally, our approach fuses the appearance feature and pose feature via element-wise addition to form the interaction representation. Experimental results show that our approach can be readily applied to existing one-stage HOI detectors. Moreover, we achieve state-of-the-art performance on two benchmarks: HICO-DET and V-COCO.
Recently, the rise of query-based Transformer decoders is reshaping camera-based 3D object detection. These query-based decoders are surpassing the traditional dense BEV (Bird's Eye View)-based methods. However, we argue that dense BEV frameworks remain important due to their outstanding abilities in depth estimation and object localization, depicting 3D scenes accurately and comprehensively. This paper aims to address the drawbacks of the existing dense BEV-based 3D object detectors by introducing our proposed enhanced components, including a CRF-modulated depth estimation module enforcing object-level consistencies, a long-term temporal aggregation module with extended receptive fields, and a two-stage object decoder combining perspective techniques with CRF-modulated depth embedding. These enhancements lead to a "modernized" dense BEV framework dubbed BEVNeXt. On the nuScenes benchmark, BEVNeXt outperforms both BEV-based and query-based frameworks under various settings, achieving a state-of-the-art result of 64.2 NDS on the nuScenes test set. Code will be available at \url{https://github.com/woxihuanjiangguo/BEVNeXt}.
Compressing large language models (LLMs), often consisting of billions of parameters, provides faster inference, smaller memory footprints, and enables local deployment. Two standard compression techniques are pruning and quantization, with the former eliminating redundant connections in model layers and the latter representing model parameters with fewer bits. The key tradeoff is between the degree of compression and the impact on the quality of the compressed model. Existing research on LLM compression primarily focuses on performance in terms of general metrics like perplexity or downstream task accuracy. More fine-grained metrics, such as those measuring parametric knowledge, remain significantly underexplored. To help bridge this gap, we present a comprehensive analysis across multiple model families (ENCODER, ENCODER-DECODER, and DECODER) using the LAMA and LM-HARNESS benchmarks in order to systematically quantify the effect of commonly employed compression techniques on model performance. A particular focus is on tradeoffs involving parametric knowledge, with the goal of providing practitioners with practical insights to help make informed decisions on compression. We release our codebase1 to enable further research.
Group Activity Recognition (GAR) is a fundamental problem in computer vision, with diverse applications in sports video analysis, video surveillance, and social scene understanding. Unlike conventional action recognition, GAR aims to classify the actions of a group of individuals as a whole, requiring a deep understanding of their interactions and spatiotemporal relationships. To address the challenges in GAR, we present REACT (\textbf{R}ecognize \textbf{E}very \textbf{Act}ion Everywhere All At Once), a novel architecture inspired by the transformer encoder-decoder model explicitly designed to model complex contextual relationships within videos, including multi-modality and spatio-temporal features. Our architecture features a cutting-edge Vision-Language Encoder block for integrated temporal, spatial, and multi-modal interaction modeling. This component efficiently encodes spatiotemporal interactions, even with sparsely sampled frames, and recovers essential local information. Our Action Decoder Block refines the joint understanding of text and video data, allowing us to precisely retrieve bounding boxes, enhancing the link between semantics and visual reality. At the core, our Actor Fusion Block orchestrates a fusion of actor-specific data and textual features, striking a balance between specificity and context. Our method outperforms state-of-the-art GAR approaches in extensive experiments, demonstrating superior accuracy in recognizing and understanding group activities. Our architecture's potential extends to diverse real-world applications, offering empirical evidence of its performance gains. This work significantly advances the field of group activity recognition, providing a robust framework for nuanced scene comprehension.
The Gaussianity assumption has been consistently criticized as a main limitation of the Variational Autoencoder (VAE) despite its efficiency in computational modeling. In this paper, we propose a new approach that expands the model capacity (i.e., expressive power of distributional family) without sacrificing the computational advantages of the VAE framework. Our VAE model's decoder is composed of an infinite mixture of asymmetric Laplace distribution, which possesses general distribution fitting capabilities for continuous variables. Our model is represented by a special form of a nonparametric M-estimator for estimating general quantile functions, and we theoretically establish the relevance between the proposed model and quantile estimation. We apply the proposed model to synthetic data generation, and particularly, our model demonstrates superiority in easily adjusting the level of data privacy.
Unsupervised object-centric learning aims to decompose scenes into interpretable object entities, termed slots. Slot-based auto-encoders stand out as a prominent method for this task. Within them, crucial aspects include guiding the encoder to generate object-specific slots and ensuring the decoder utilizes them during reconstruction. This work introduces two novel techniques, (i) an attention-based self-training approach, which distills superior slot-based attention masks from the decoder to the encoder, enhancing object segmentation, and (ii) an innovative patch-order permutation strategy for autoregressive transformers that strengthens the role of slot vectors in reconstruction. The effectiveness of these strategies is showcased experimentally. The combined approach significantly surpasses prior slot-based autoencoder methods in unsupervised object segmentation, especially with complex real-world images. We provide the implementation code at https://github.com/gkakogeorgiou/spot .
Multimodal (e.g., RGB-Depth/RGB-Thermal) fusion has shown great potential for improving semantic segmentation in complex scenes (e.g., indoor/low-light conditions). Existing approaches often fully fine-tune a dual-branch encoder-decoder framework with a complicated feature fusion strategy for achieving multimodal semantic segmentation, which is training-costly due to the massive parameter updates in feature extraction and fusion. To address this issue, we propose a surprisingly simple yet effective dual-prompt learning network (dubbed DPLNet) for training-efficient multimodal (e.g., RGB-D/T) semantic segmentation. The core of DPLNet is to directly adapt a frozen pre-trained RGB model to multimodal semantic segmentation, reducing parameter updates. For this purpose, we present two prompt learning modules, comprising multimodal prompt generator (MPG) and multimodal feature adapter (MFA). MPG works to fuse the features from different modalities in a compact manner and is inserted from shadow to deep stages to generate the multi-level multimodal prompts that are injected into the frozen backbone, while MPG adapts prompted multimodal features in the frozen backbone for better multimodal semantic segmentation. Since both the MPG and MFA are lightweight, only a few trainable parameters (3.88M, 4.4% of the pre-trained backbone parameters) are introduced for multimodal feature fusion and learning. Using a simple decoder (3.27M parameters), DPLNet achieves new state-of-the-art performance or is on a par with other complex approaches on four RGB-D/T semantic segmentation datasets while satisfying parameter efficiency. Moreover, we show that DPLNet is general and applicable to other multimodal tasks such as salient object detection and video semantic segmentation. Without special design, DPLNet outperforms many complicated models. Our code will be available at github.com/ShaohuaDong2021/DPLNet.
Climate change has led to an increase in frequency of extreme weather events. Early warning systems can prevent disasters and loss of life. Managing such events remain a challenge for both public and private institutions. Precipitation nowcasting can help relevant institutions to better prepare for such events. Numerical weather prediction (NWP) has traditionally been used to make physics based forecasting, and recently deep learning based approaches have been used to reduce turn-around time for nowcasting. In this work, recently proposed Swin-UNETR (Swin UNEt TRansformer) is used for precipitation nowcasting for ten different regions of Europe. Swin-UNETR utilizes a U-shaped network within which a swin transformer-based encoder extracts multi-scale features from multiple input channels of satellite image, while CNN-based decoder makes the prediction. Trained model is capable of nowcasting not only for the regions for which data is available, but can also be used for new regions for which data is not available.
Human-object interaction (HOI) detection aims to comprehend the intricate relationships between humans and objects, predicting $<human, action, object>$ triplets, and serving as the foundation for numerous computer vision tasks. The complexity and diversity of human-object interactions in the real world, however, pose significant challenges for both annotation and recognition, particularly in recognizing interactions within an open world context. This study explores the universal interaction recognition in an open-world setting through the use of Vision-Language (VL) foundation models and large language models (LLMs). The proposed method is dubbed as \emph{\textbf{UniHOI}}. We conduct a deep analysis of the three hierarchical features inherent in visual HOI detectors and propose a method for high-level relation extraction aimed at VL foundation models, which we call HO prompt-based learning. Our design includes an HO Prompt-guided Decoder (HOPD), facilitates the association of high-level relation representations in the foundation model with various HO pairs within the image. Furthermore, we utilize a LLM (\emph{i.e.} GPT) for interaction interpretation, generating a richer linguistic understanding for complex HOIs. For open-category interaction recognition, our method supports either of two input types: interaction phrase or interpretive sentence. Our efficient architecture design and learning methods effectively unleash the potential of the VL foundation models and LLMs, allowing UniHOI to surpass all existing methods with a substantial margin, under both supervised and zero-shot settings. The code and pre-trained weights are available at: \url{https://github.com/Caoyichao/UniHOI}.
A vital and rapidly growing application, remote sensing offers vast yet sparsely labeled, spatially aligned multimodal data; this makes self-supervised learning algorithms invaluable. We present CROMA: a framework that combines contrastive and reconstruction self-supervised objectives to learn rich unimodal and multimodal representations. Our method separately encodes masked-out multispectral optical and synthetic aperture radar samples -- aligned in space and time -- and performs cross-modal contrastive learning. Another encoder fuses these sensors, producing joint multimodal encodings that are used to predict the masked patches via a lightweight decoder. We show that these objectives are complementary when leveraged on spatially aligned multimodal data. We also introduce X- and 2D-ALiBi, which spatially biases our cross- and self-attention matrices. These strategies improve representations and allow our models to effectively extrapolate to images up to 17.6x larger at test-time. CROMA outperforms the current SoTA multispectral model, evaluated on: four classification benchmarks -- finetuning (avg. 1.8%), linear (avg. 2.4%) and nonlinear (avg. 1.4%) probing, kNN classification (avg. 3.5%), and K-means clustering (avg. 8.4%); and three segmentation benchmarks (avg. 6.4%). CROMA's rich, optionally multimodal representations can be widely leveraged across remote sensing applications.
Reconstructing 3D clothed human avatars from single images is a challenging task, especially when encountering complex poses and loose clothing. Current methods exhibit limitations in performance, largely attributable to their dependence on insufficient 2D image features and inconsistent query methods. Owing to this, we present the Global-correlated 3D-decoupling Transformer for clothed Avatar reconstruction (GTA), a novel transformer-based architecture that reconstructs clothed human avatars from monocular images. Our approach leverages transformer architectures by utilizing a Vision Transformer model as an encoder for capturing global-correlated image features. Subsequently, our innovative 3D-decoupling decoder employs cross-attention to decouple tri-plane features, using learnable embeddings as queries for cross-plane generation. To effectively enhance feature fusion with the tri-plane 3D feature and human body prior, we propose a hybrid prior fusion strategy combining spatial and prior-enhanced queries, leveraging the benefits of spatial localization and human body prior knowledge. Comprehensive experiments on CAPE and THuman2.0 datasets illustrate that our method outperforms state-of-the-art approaches in both geometry and texture reconstruction, exhibiting high robustness to challenging poses and loose clothing, and producing higher-resolution textures. Codes will be available at https://github.com/River-Zhang/GTA.
Masked image modeling (MIM) has become a prevalent pre-training setup for vision foundation models and attains promising performance. Despite its success, existing MIM methods discard the decoder network during downstream applications, resulting in inconsistent representations between pre-training and fine-tuning and can hamper downstream task performance. In this paper, we propose a new architecture, RevColV2, which tackles this issue by keeping the entire autoencoder architecture during both pre-training and fine-tuning. The main body of RevColV2 contains bottom-up columns and top-down columns, between which information is reversibly propagated and gradually disentangled. Such design enables our architecture with the nice property: maintaining disentangled low-level and semantic information at the end of the network in MIM pre-training. Our experimental results suggest that a foundation model with decoupled features can achieve competitive performance across multiple downstream vision tasks such as image classification, semantic segmentation and object detection. For example, after intermediate fine-tuning on ImageNet-22K dataset, RevColV2-L attains 88.4% top-1 accuracy on ImageNet-1K classification and 58.6 mIoU on ADE20K semantic segmentation. With extra teacher and large scale dataset, RevColv2-L achieves 62.1 box AP on COCO detection and 60.4 mIoU on ADE20K semantic segmentation. Code and models are released at https://github.com/megvii-research/RevCol
This paper outlines an end-to-end optimized lossy image compression framework using diffusion generative models. The approach relies on the transform coding paradigm, where an image is mapped into a latent space for entropy coding and, from there, mapped back to the data space for reconstruction. In contrast to VAE-based neural compression, where the (mean) decoder is a deterministic neural network, our decoder is a conditional diffusion model. Our approach thus introduces an additional ``content'' latent variable on which the reverse diffusion process is conditioned and uses this variable to store information about the image. The remaining ``texture'' variables characterizing the diffusion process are synthesized at decoding time. We show that the model's performance can be tuned toward perceptual metrics of interest. Our extensive experiments involving multiple datasets and image quality assessment metrics show that our approach yields stronger reported FID scores than the GAN-based model, while also yielding competitive performance with VAE-based models in several distortion metrics. Furthermore, training the diffusion with $\mathcal{X}$-parameterization enables high-quality reconstructions in only a handful of decoding steps, greatly affecting the model's practicality. Our code is available at: \url{https://github.com/buggyyang/CDC_compression}
Prompting is the primary way to utilize the multitask capabilities of language models (LMs), but prompts occupy valuable space in the input context window, and repeatedly encoding the same prompt is computationally inefficient. Finetuning and distillation methods allow for specialization of LMs without prompting, but require retraining the model for each task. To avoid this trade-off entirely, we present gisting, which trains an LM to compress prompts into smaller sets of "gist" tokens which can be cached and reused for compute efficiency. Gist models can be trained with no additional cost over standard instruction finetuning by simply modifying Transformer attention masks to encourage prompt compression. On decoder (LLaMA-7B) and encoder-decoder (FLAN-T5-XXL) LMs, gisting enables up to 26x compression of prompts, resulting in up to 40% FLOPs reductions, 4.2% wall time speedups, and storage savings, all with minimal loss in output quality.
In the framework of learned image compression, the context model plays a pivotal role in capturing the dependencies among latent representations. To reduce the decoding time resulting from the serial autoregressive context model, the parallel context model has been proposed as an alternative that necessitates only two passes during the decoding phase, thus facilitating efficient image compression in real-world scenarios. However, performance degradation occurs due to its incomplete casual context. To tackle this issue, we conduct an in-depth analysis of the performance degradation observed in existing parallel context models, focusing on two aspects: the Quantity and Quality of information utilized for context prediction and decoding. Based on such analysis, we propose the \textbf{Corner-to-Center transformer-based Context Model (C$^3$M)} designed to enhance context and latent predictions and improve rate-distortion performance. Specifically, we leverage the logarithmic-based prediction order to predict more context features from corner to center progressively. In addition, to enlarge the receptive field in the analysis and synthesis transformation, we use the Long-range Crossing Attention Module (LCAM) in the encoder/decoder to capture the long-range semantic information by assigning the different window shapes in different channels. Extensive experimental evaluations show that the proposed method is effective and outperforms the state-of-the-art parallel methods. Finally, according to the subjective analysis, we suggest that improving the detailed representation in transformer-based image compression is a promising direction to be explored.
Autoregressive transformers are spectacular models for short sequences but scale poorly to long sequences such as high-resolution images, podcasts, code, or books. We proposed Megabyte, a multi-scale decoder architecture that enables end-to-end differentiable modeling of sequences of over one million bytes. Megabyte segments sequences into patches and uses a local submodel within patches and a global model between patches. This enables sub-quadratic self-attention, much larger feedforward layers for the same compute, and improved parallelism during decoding -- unlocking better performance at reduced cost for both training and generation. Extensive experiments show that Megabyte allows byte-level models to perform competitively with subword models on long context language modeling, achieve state-of-the-art density estimation on ImageNet, and model audio from raw files. Together, these results establish the viability of tokenization-free autoregressive sequence modeling at scale.