We propose a multi-layer variational autoencoder method, we call HR-VQVAE, that learns hierarchical discrete representations of the data. By utilizing a novel objective function, each layer in HR-VQVAE learns a discrete representation of the residual from previous layers through a vector quantized encoder. Furthermore, the representations at each layer are hierarchically linked to those at previous layers. We evaluate our method on the tasks of image reconstruction and generation. Experimental results demonstrate that the discrete representations learned by HR-VQVAE enable the decoder to reconstruct high-quality images with less distortion than the baseline methods, namely VQVAE and VQVAE-2. HR-VQVAE can also generate high-quality and diverse images that outperform state-of-the-art generative models, providing further verification of the efficiency of the learned representations. The hierarchical nature of HR-VQVAE i) reduces the decoding search time, making the method particularly suitable for high-load tasks and ii) allows to increase the codebook size without incurring the codebook collapse problem.
A system that could reliably identify and sum up the most important points of a conversation would be valuable in a wide variety of real-world contexts, from business meetings to medical consultations to customer service calls. Recent advances in deep learning, and especially the invention of encoder-decoder architectures, has significantly improved language generation systems, opening the door to improved forms of abstractive summarization, a form of summarization particularly well-suited for multi-party conversation. In this paper, we provide an overview of the challenges raised by the task of abstractive meeting summarization and of the data sets, models and evaluation metrics that have been used to tackle the problems.
Despite advances in generating fluent texts, existing pretraining models tend to attach incoherent event sequences to involved entities when generating narratives such as stories and news. We conjecture that such issues result from representing entities as static embeddings of superficial words, while neglecting to model their ever-changing states, i.e., the information they carry, as the text unfolds. Therefore, we extend the Transformer model to dynamically conduct entity state updates and sentence realization for narrative generation. We propose a contrastive framework to learn the state representations in a discrete space, and insert additional attention layers into the decoder to better exploit these states. Experiments on two narrative datasets show that our model can generate more coherent and diverse narratives than strong baselines with the guidance of meaningful entity states.
Video recognition has been dominated by the end-to-end learning paradigm -- first initializing a video recognition model with weights of a pretrained image model and then conducting end-to-end training on videos. This enables the video network to benefit from the pretrained image model. However, this requires substantial computation and memory resources for finetuning on videos and the alternative of directly using pretrained image features without finetuning the image backbone leads to subpar results. Fortunately, recent advances in Contrastive Vision-Language Pre-training (CLIP) pave the way for a new route for visual recognition tasks. Pretrained on large open-vocabulary image-text pair data, these models learn powerful visual representations with rich semantics. In this paper, we present Efficient Video Learning (EVL) -- an efficient framework for directly training high-quality video recognition models with frozen CLIP features. Specifically, we employ a lightweight Transformer decoder and learn a query token to dynamically collect frame-level spatial features from the CLIP image encoder. Furthermore, we adopt a local temporal module in each decoder layer to discover temporal clues from adjacent frames and their attention maps. We show that despite being efficient to train with a frozen backbone, our models learn high quality video representations on a variety of video recognition datasets. Code is available at https://github.com/OpenGVLab/efficient-video-recognition.
In an unsupervised attack on variational autoencoders (VAEs), an adversary finds a small perturbation in an input sample that significantly changes its latent space encoding, thereby compromising the reconstruction for a fixed decoder. A known reason for such vulnerability is the distortions in the latent space resulting from a mismatch between approximated latent posterior and a prior distribution. Consequently, a slight change in an input sample can move its encoding to a low/zero density region in the latent space resulting in an unconstrained generation. This paper demonstrates that an optimal way for an adversary to attack VAEs is to exploit a directional bias of a stochastic pullback metric tensor induced by the encoder and decoder networks. The pullback metric tensor of an encoder measures the change in infinitesimal latent volume from an input to a latent space. Thus, it can be viewed as a lens to analyse the effect of input perturbations leading to latent space distortions. We propose robustness evaluation scores using the eigenspectrum of a pullback metric tensor. Moreover, we empirically show that the scores correlate with the robustness parameter $\beta$ of the $\beta-$VAE. Since increasing $\beta$ also degrades reconstruction quality, we demonstrate a simple alternative using \textit{mixup} training to fill the empty regions in the latent space, thus improving robustness with improved reconstruction.
Image restoration algorithms such as super resolution (SR) are indispensable pre-processing modules for object detection in low quality images. Most of these algorithms assume the degradation is fixed and known a priori. However, in practical, either the real degradation or optimal up-sampling ratio rate is unknown or differs from assumption, leading to a deteriorating performance for both the pre-processing module and the consequent high-level task such as object detection. Here, we propose a novel self-supervised framework to detect objects in degraded low resolution images. We utilizes the downsampling degradation as a kind of transformation for self-supervised signals to explore the equivariant representation against various resolutions and other degradation conditions. The Auto Encoding Resolution in Self-supervision (AERIS) framework could further take the advantage of advanced SR architectures with an arbitrary resolution restoring decoder to reconstruct the original correspondence from the degraded input image. Both the representation learning and object detection are optimized jointly in an end-to-end training fashion. The generic AERIS framework could be implemented on various mainstream object detection architectures with different backbones. The extensive experiments show that our methods has achieved superior performance compared with existing methods when facing variant degradation situations. Code would be released at https://github.com/cuiziteng/ECCV_AERIS.
Image matting refers to predicting the alpha values of unknown foreground areas from natural images. Prior methods have focused on propagating alpha values from known to unknown regions. However, not all natural images have a specifically known foreground. Images of transparent objects, like glass, smoke, web, etc., have less or no known foreground. In this paper, we propose a Transformer-based network, TransMatting, to model transparent objects with a big receptive field. Specifically, we redesign the trimap as three learnable tri-tokens for introducing advanced semantic features into the self-attention mechanism. A small convolutional network is proposed to utilize the global feature and non-background mask to guide the multi-scale feature propagation from encoder to decoder for maintaining the contexture of transparent objects. In addition, we create a high-resolution matting dataset of transparent objects with small known foreground areas. Experiments on several matting benchmarks demonstrate the superiority of our proposed method over the current state-of-the-art methods.
High-confidence overlap prediction and accurate correspondences are critical for cutting-edge models to align paired point clouds in a partial-to-partial manner. However, there inherently exists uncertainty between the overlapping and non-overlapping regions, which has always been neglected and significantly affects the registration performance. Beyond the current wisdom, we propose a novel uncertainty-aware overlap prediction network, dubbed UTOPIC, to tackle the ambiguous overlap prediction problem; to our knowledge, this is the first to explicitly introduce overlap uncertainty to point cloud registration. Moreover, we induce the feature extractor to implicitly perceive the shape knowledge through a completion decoder, and present a geometric relation embedding for Transformer to obtain transformation-invariant geometry-aware feature representations. With the merits of more reliable overlap scores and more precise dense correspondences, UTOPIC can achieve stable and accurate registration results, even for the inputs with limited overlapping areas. Extensive quantitative and qualitative experiments on synthetic and real benchmarks demonstrate the superiority of our approach over state-of-the-art methods.
For a monocular 360 image, depth estimation is a challenging because the distortion increases along the latitude. To perceive the distortion, existing methods devote to designing a deep and complex network architecture. In this paper, we provide a new perspective that constructs an interpretable and sparse representation for a 360 image. Considering the importance of the geometric structure in depth estimation, we utilize the contourlet transform to capture an explicit geometric cue in the spectral domain and integrate it with an implicit cue in the spatial domain. Specifically, we propose a neural contourlet network consisting of a convolutional neural network and a contourlet transform branch. In the encoder stage, we design a spatial-spectral fusion module to effectively fuse two types of cues. Contrary to the encoder, we employ the inverse contourlet transform with learned low-pass subbands and band-pass directional subbands to compose the depth in the decoder. Experiments on the three popular panoramic image datasets demonstrate that the proposed approach outperforms the state-of-the-art schemes with faster convergence. Code is available at https://github.com/zhijieshen-bjtu/Neural-Contourlet-Network-for-MODE.
Learned image compression allows achieving state-of-the-art accuracy and compression ratios, but their relatively slow runtime performance limits their usage. While previous attempts on optimizing learned image codecs focused more on the neural model and entropy coding, we present an alternative method to improving the runtime performance of various learned image compression models. We introduce multi-threaded pipelining and an optimized memory model to enable GPU and CPU workloads asynchronous execution, fully taking advantage of computational resources. Our architecture alone already produces excellent performance without any change to the neural model itself. We also demonstrate that combining our architecture with previous tweaks to the neural models can further improve runtime performance. We show that our implementations excel in throughput and latency compared to the baseline and demonstrate the performance of our implementations by creating a real-time video streaming encoder-decoder sample application, with the encoder running on an embedded device.
It is well believed that Transformer performs better in semantic segmentation compared to convolutional neural networks. Nevertheless, the original Vision Transformer may lack of inductive biases of local neighborhoods and possess a high time complexity. Recently, Swin Transformer sets a new record in various vision tasks by using hierarchical architecture and shifted windows while being more efficient. However, as Swin Transformer is specifically designed for image classification, it may achieve suboptimal performance on dense prediction-based segmentation task. Further, simply combing Swin Transformer with existing methods would lead to the boost of model size and parameters for the final segmentation model. In this paper, we rethink the Swin Transformer for semantic segmentation, and design a lightweight yet effective transformer model, called SSformer. In this model, considering the inherent hierarchical design of Swin Transformer, we propose a decoder to aggregate information from different layers, thus obtaining both local and global attentions. Experimental results show the proposed SSformer yields comparable mIoU performance with state-of-the-art models, while maintaining a smaller model size and lower compute.
The automatic segmentation of blood vessels in fundus images can help analyze the condition of retinal vasculature, which is crucial for identifying various systemic diseases like hypertension, diabetes, etc. Despite the success of Deep Learning-based models in this segmentation task, most of them are heavily parametrized and thus have limited use in practical applications. This paper proposes IterMiUnet, a new lightweight convolution-based segmentation model that requires significantly fewer parameters and yet delivers performance similar to existing models. The model makes use of the excellent segmentation capabilities of Iternet architecture but overcomes its heavily parametrized nature by incorporating the encoder-decoder structure of MiUnet model within it. Thus, the new model reduces parameters without any compromise with the network's depth, which is necessary to learn abstract hierarchical concepts in deep models. This lightweight segmentation model speeds up training and inference time and is potentially helpful in the medical domain where data is scarce and, therefore, heavily parametrized models tend to overfit. The proposed model was evaluated on three publicly available datasets: DRIVE, STARE, and CHASE-DB1. Further cross-training and inter-rater variability evaluations have also been performed. The proposed model has a lot of potential to be utilized as a tool for the early diagnosis of many diseases.
Transformer-based pretrained language models (LMs) are ubiquitous across natural language understanding, but cannot be applied to long sequences such as stories, scientific articles and long documents, due to their quadratic complexity. While a myriad of efficient transformer variants have been proposed, they are typically based on custom implementations that require expensive pretraining from scratch. In this work, we propose SLED: SLiding-Encoder and Decoder, a simple approach for processing long sequences that re-uses and leverages battle-tested short-text pretrained LMs. Specifically, we partition the input into overlapping chunks, encode each with a short-text LM encoder and use the pretrained decoder to fuse information across chunks (fusion-in-decoder). We illustrate through controlled experiments that SLED offers a viable strategy for long text understanding and evaluate our approach on SCROLLS, a benchmark with seven datasets across a wide range of language understanding tasks. We find that SLED is competitive with specialized models that are up to 50x larger and require a dedicated and expensive pretraining step.
Transformer structure, stacked by a sequence of encoder and decoder network layers, achieves significant development in neural machine translation. However, vanilla Transformer mainly exploits the top-layer representation, assuming the lower layers provide trivial or redundant information and thus ignoring the bottom-layer feature that is potentially valuable. In this work, we propose the Group-Transformer model (GTrans) that flexibly divides multi-layer representations of both encoder and decoder into different groups and then fuses these group features to generate target words. To corroborate the effectiveness of the proposed method, extensive experiments and analytic experiments are conducted on three bilingual translation benchmarks and two multilingual translation tasks, including the IWLST-14, IWLST-17, LDC, WMT-14 and OPUS-100 benchmark. Experimental and analytical results demonstrate that our model outperforms its Transformer counterparts by a consistent gain. Furthermore, it can be successfully scaled up to 60 encoder layers and 36 decoder layers.
We show that pre-trained Generative Adversarial Networks (GANs) such as StyleGAN and BigGAN can be used as a latent bank to improve the performance of image super-resolution. While most existing perceptual-oriented approaches attempt to generate realistic outputs through learning with adversarial loss, our method, Generative LatEnt bANk (GLEAN), goes beyond existing practices by directly leveraging rich and diverse priors encapsulated in a pre-trained GAN. But unlike prevalent GAN inversion methods that require expensive image-specific optimization at runtime, our approach only needs a single forward pass for restoration. GLEAN can be easily incorporated in a simple encoder-bank-decoder architecture with multi-resolution skip connections. Employing priors from different generative models allows GLEAN to be applied to diverse categories (\eg~human faces, cats, buildings, and cars). We further present a lightweight version of GLEAN, named LightGLEAN, which retains only the critical components in GLEAN. Notably, LightGLEAN consists of only 21% of parameters and 35% of FLOPs while achieving comparable image quality. We extend our method to different tasks including image colorization and blind image restoration, and extensive experiments show that our proposed models perform favorably in comparison to existing methods. Codes and models are available at https://github.com/open-mmlab/mmediting.
In this work, we propose a fully differentiable graph neural network (GNN)-based architecture for channel decoding and showcase a competitive decoding performance for various coding schemes, such as low-density parity-check (LDPC) and BCH codes. The idea is to let a neural network (NN) learn a generalized message passing algorithm over a given graph that represents the forward error correction (FEC) code structure by replacing node and edge message updates with trainable functions. Contrary to many other deep learning-based decoding approaches, the proposed solution enjoys scalability to arbitrary block lengths and the training is not limited by the curse of dimensionality. We benchmark our proposed decoder against state-of-the-art in conventional channel decoding as well as against recent deep learning-based results. For the (63,45) BCH code, our solution outperforms weighted belief propagation (BP) decoding by approximately 0.4 dB with significantly less decoding iterations and even for 5G NR LDPC codes, we observe a competitive performance when compared to conventional BP decoding. For the BCH codes, the resulting GNN decoder can be fully parametrized with only 9640 weights.
Existing learning-based image inpainting methods are still in challenge when facing complex semantic environments and diverse hole patterns. The prior information learned from the large scale training data is still insufficient for these situations. Reference images captured covering the same scenes share similar texture and structure priors with the corrupted images, which offers new prospects for the image inpainting tasks. Inspired by this, we first build a benchmark dataset containing 10K pairs of input and reference images for reference-guided inpainting. Then we adopt an encoder-decoder structure to separately infer the texture and structure features of the input image considering their pattern discrepancy of texture and structure during inpainting. A feature alignment module is further designed to refine these features of the input image with the guidance of a reference image. Both quantitative and qualitative evaluations demonstrate the superiority of our method over the state-of-the-art methods in terms of completing complex holes.
Transformer encoder architectures have recently achieved state-of-the-art results on monocular 3D human mesh reconstruction, but they require a substantial number of parameters and expensive computations. Due to the large memory overhead and slow inference speed, it is difficult to deploy such models for practical use. In this paper, we propose a novel transformer encoder-decoder architecture for 3D human mesh reconstruction from a single image, called FastMETRO. We identify the performance bottleneck in the encoder-based transformers is caused by the token design which introduces high complexity interactions among input tokens. We disentangle the interactions via an encoder-decoder architecture, which allows our model to demand much fewer parameters and shorter inference time. In addition, we impose the prior knowledge of human body's morphological relationship via attention masking and mesh upsampling operations, which leads to faster convergence with higher accuracy. Our FastMETRO improves the Pareto-front of accuracy and efficiency, and clearly outperforms image-based methods on Human3.6M and 3DPW. Furthermore, we validate its generalizability on FreiHAND.
Masked image modeling (MIM) has achieved promising results on various vision tasks. However, the limited discriminability of learned representation manifests there is still plenty to go for making a stronger vision learner. Towards this goal, we propose Contrastive Masked Autoencoders (CMAE), a new self-supervised pre-training method for learning more comprehensive and capable vision representations. By elaboratively unifying contrastive learning (CL) and masked image model (MIM) through novel designs, CMAE leverages their respective advantages and learns representations with both strong instance discriminability and local perceptibility. Specifically, CMAE consists of two branches where the online branch is an asymmetric encoder-decoder and the momentum branch is a momentum updated encoder. During training, the online encoder reconstructs original images from latent representations of masked images to learn holistic features. The momentum encoder, fed with the full images, enhances the feature discriminability via contrastive learning with its online counterpart. To make CL compatible with MIM, CMAE introduces two new components, i.e. pixel shifting for generating plausible positive views and feature decoder for complementing features of contrastive pairs. Thanks to these novel designs, CMAE effectively improves the representation quality and transfer performance over its MIM counterpart. CMAE achieves the state-of-the-art performance on highly competitive benchmarks of image classification, semantic segmentation and object detection. Notably, CMAE-Base achieves $85.3\%$ top-1 accuracy on ImageNet and $52.5\%$ mIoU on ADE20k, surpassing previous best results by $0.7\%$ and $1.8\%$ respectively. The source code is publicly accessible at \url{https://github.com/ZhichengHuang/CMAE}.
Detection transformer (DETR) relies on one-to-one assignment, assigning one ground-truth object to one prediction, for end-to-end detection without NMS post-processing. It is known that one-to-many assignment, assigning one ground-truth object to multiple predictions, succeeds in detection methods such as Faster R-CNN and FCOS. While the naive one-to-many assignment does not work for DETR, and it remains challenging to apply one-to-many assignment for DETR training. In this paper, we introduce Group DETR, a simple yet efficient DETR training approach that introduces a group-wise way for one-to-many assignment. This approach involves using multiple groups of object queries, conducting one-to-one assignment within each group, and performing decoder self-attention separately. It resembles data augmentation with automatically-learned object query augmentation. It is also equivalent to simultaneously training parameter-sharing networks of the same architecture, introducing more supervision and thus improving DETR training. The inference process is the same as DETR trained normally and only needs one group of queries without any architecture modification. Group DETR is versatile and is applicable to various DETR variants. The experiments show that Group DETR significantly speeds up the training convergence and improves the performance of various DETR-based models. Code will be available at \url{https://github.com/Atten4Vis/GroupDETR}.