In this paper, we propose a Vision-Audio-Language Omni-peRception pretraining model (VALOR) for multi-modal understanding and generation. Different from widely-studied vision-language pretraining models, VALOR jointly models relationships of vision, audio and language in an end-to-end manner. It contains three separate encoders for single modality representations, and a decoder for multimodal conditional text generation. We design two pretext tasks to pretrain VALOR model, including Multimodal Grouping Alignment (MGA) and Multimodal Grouping Captioning (MGC). MGA projects vision, language and audio to the same common space, building vision-language, audio-language and audiovisual-language alignment simultaneously. MGC learns how to generate text tokens in conditions of vision, audio or their both. To promote vision-audio-language pretraining research, we construct a large-scale high-quality tri-modality dataset named VALOR-1M, which contains 1M audiable videos with human annotated audiovisual captions. Extensive experiments show that VALOR can learn strong multimodal correlations and be generalized to various downstream tasks (e.g., retrieval, captioning and question answering), with different input modalities (e.g., vision-language, audio-language and audiovisual-language). VALOR achieves new state-of-the-art performances on series of public cross-modality benchmarks. Code and data are available at project page https://casia-iva-group.github.io/projects/VALOR.
Semantic Scene Completion (SSC) transforms an image of single-view depth and/or RGB 2D pixels into 3D voxels, each of whose semantic labels are predicted. SSC is a well-known ill-posed problem as the prediction model has to "imagine" what is behind the visible surface, which is usually represented by Truncated Signed Distance Function (TSDF). Due to the sensory imperfection of the depth camera, most existing methods based on the noisy TSDF estimated from depth values suffer from 1) incomplete volumetric predictions and 2) confused semantic labels. To this end, we use the ground-truth 3D voxels to generate a perfect visible surface, called TSDF-CAD, and then train a "cleaner" SSC model. As the model is noise-free, it is expected to focus more on the "imagination" of unseen voxels. Then, we propose to distill the intermediate "cleaner" knowledge into another model with noisy TSDF input. In particular, we use the 3D occupancy feature and the semantic relations of the "cleaner self" to supervise the counterparts of the "noisy self" to respectively address the above two incorrect predictions. Experimental results validate that our method improves the noisy counterparts with 3.1% IoU and 2.2% mIoU for measuring scene completion and SSC, and also achieves new state-of-the-art accuracy on the popular NYU dataset.
Although numerous solutions have been proposed for image super-resolution, they are usually incompatible with low-power devices with many computational and memory constraints. In this paper, we address this problem by proposing a simple yet effective deep network to solve image super-resolution efficiently. In detail, we develop a spatially-adaptive feature modulation (SAFM) mechanism upon a vision transformer (ViT)-like block. Within it, we first apply the SAFM block over input features to dynamically select representative feature representations. As the SAFM block processes the input features from a long-range perspective, we further introduce a convolutional channel mixer (CCM) to simultaneously extract local contextual information and perform channel mixing. Extensive experimental results show that the proposed method is $3\times$ smaller than state-of-the-art efficient SR methods, e.g., IMDN, in terms of the network parameters and requires less computational cost while achieving comparable performance. The code is available at https://github.com/sunny2109/SAFMN.
Part feature learning is a critical technology for finegrained semantic understanding in vehicle re-identification. However, recent unsupervised re-identification works exhibit serious gradient collapse issues when directly modeling the part features and global features. To address this problem, in this paper, we propose a novel Triplet Contrastive Learning framework (TCL) which leverages cluster features to bridge the part features and global features. Specifically, TCL devises three memory banks to store the features according to their attributes and proposes a proxy contrastive loss (PCL) to make contrastive learning between adjacent memory banks, thus presenting the associations between the part and global features as a transition of the partcluster and cluster-global associations. Since the cluster memory bank deals with all the instance features, it can summarize them into a discriminative feature representation. To deeply exploit the instance information, TCL proposes two additional loss functions. For the inter-class instance, a hybrid contrastive loss (HCL) re-defines the sample correlations by approaching the positive cluster features and leaving the all negative instance features. For the intra-class instances, a weighted regularization cluster contrastive loss (WRCCL) refines the pseudo labels by penalizing the mislabeled images according to the instance similarity. Extensive experiments show that TCL outperforms many state-of-the-art unsupervised vehicle re-identification approaches. The code will be available at https://github.com/muzishen/TCL.
We propose an effective lightweight dynamic local and global self-attention network (DLGSANet) to solve image super-resolution. Our method explores the properties of Transformers while having low computational costs. Motivated by the network designs of Transformers, we develop a simple yet effective multi-head dynamic local self-attention (MHDLSA) module to extract local features efficiently. In addition, we note that existing Transformers usually explore all similarities of the tokens between the queries and keys for the feature aggregation. However, not all the tokens from the queries are relevant to those in keys, using all the similarities does not effectively facilitate the high-resolution image reconstruction. To overcome this problem, we develop a sparse global self-attention (SparseGSA) module to select the most useful similarity values so that the most useful global features can be better utilized for the high-resolution image reconstruction. We develop a hybrid dynamic-Transformer block(HDTB) that integrates the MHDLSA and SparseGSA for both local and global feature exploration. To ease the network training, we formulate the HDTBs into a residual hybrid dynamic-Transformer group (RHDTG). By embedding the RHDTGs into an end-to-end trainable network, we show that our proposed method has fewer network parameters and lower computational costs while achieving competitive performance against state-of-the-art ones in terms of accuracy. More information is available at https://neonleexiang.github.io/DLGSANet/
How to effectively explore the colors of reference exemplars and propagate them to colorize each frame is vital for exemplar-based video colorization. In this paper, we present an effective BiSTNet to explore colors of reference exemplars and utilize them to help video colorization by a bidirectional temporal feature fusion with the guidance of semantic image prior. We first establish the semantic correspondence between each frame and the reference exemplars in deep feature space to explore color information from reference exemplars. Then, to better propagate the colors of reference exemplars into each frame and avoid the inaccurate matches colors from exemplars we develop a simple yet effective bidirectional temporal feature fusion module to better colorize each frame. We note that there usually exist color-bleeding artifacts around the boundaries of the important objects in videos. To overcome this problem, we further develop a mixed expert block to extract semantic information for modeling the object boundaries of frames so that the semantic image prior can better guide the colorization process for better performance. In addition, we develop a multi-scale recurrent block to progressively colorize frames in a coarse-to-fine manner. Extensive experimental results demonstrate that the proposed BiSTNet performs favorably against state-of-the-art methods on the benchmark datasets. Our code will be made available at \url{https://yyang181.github.io/BiSTNet/}
How to effectively explore semantic feature is vital for low-light image enhancement (LLE). Existing methods usually utilize the semantic feature that is only drawn from the semantic map produced by high-level semantic segmentation network (SSN). However, if the semantic map is not accurately estimated, it would affect the high-level semantic feature (HSF) extraction, which accordingly interferes with LLE. In this paper, we develop a simple yet effective two-branch semantic-aware LLE network (SLLEN) that neatly integrates the random intermediate embedding feature (IEF) (i.e., the information extracted from the intermediate layer of semantic segmentation network) together with the HSF into a unified framework for better LLE. Specifically, for one branch, we utilize an attention mechanism to integrate HSF into low-level feature. For the other branch, we extract IEF to guide the adjustment of low-level feature using nonlinear transformation manner. Finally, semantic-aware features obtained from two branches are fused and decoded for image enhancement. It is worth mentioning that IEF has some randomness compared to HSF despite their similarity on semantic characteristics, thus its introduction can allow network to learn more possibilities by leveraging the latent relationships between the low-level feature and semantic feature, just like the famous saying "God rolls the dice" in Physics Nobel Prize 2022. Comparisons between the proposed SLLEN and other state-of-the-art techniques demonstrate the superiority of SLLEN with respect to LLE quality over all the comparable alternatives.
TIReID aims to retrieve the image corresponding to the given text query from a pool of candidate images. Existing methods employ prior knowledge from single-modality pre-training to facilitate learning, but lack multi-modal correspondences. Besides, due to the substantial gap between modalities, existing methods embed the original modal features into the same latent space for cross-modal alignment. However, feature embedding may lead to intra-modal information distortion. Recently, CLIP has attracted extensive attention from researchers due to its powerful semantic concept learning capacity and rich multi-modal knowledge, which can help us solve the above problems. Accordingly, in the paper, we propose a CLIP-driven Fine-grained information excavation framework (CFine) to fully utilize the powerful knowledge of CLIP for TIReID. To transfer the multi-modal knowledge effectively, we perform fine-grained information excavation to mine intra-modal discriminative clues and inter-modal correspondences. Specifically, we first design a multi-grained global feature learning module to fully mine intra-modal discriminative local information, which can emphasize identity-related discriminative clues by enhancing the interactions between global image (text) and informative local patches (words). Secondly, cross-grained feature refinement (CFR) and fine-grained correspondence discovery (FCD) modules are proposed to establish the cross-grained and fine-grained interactions between modalities, which can filter out non-modality-shared image patches/words and mine cross-modal correspondences from coarse to fine. CFR and FCD are removed during inference to save computational costs. Note that the above process is performed in the original modality space without further feature embedding. Extensive experiments on multiple benchmarks demonstrate the superior performance of our method on TIReID.
Visual feature pyramid has shown its superiority in both effectiveness and efficiency in a wide range of applications. However, the existing methods exorbitantly concentrate on the inter-layer feature interactions but ignore the intra-layer feature regulations, which are empirically proved beneficial. Although some methods try to learn a compact intra-layer feature representation with the help of the attention mechanism or the vision transformer, they ignore the neglected corner regions that are important for dense prediction tasks. To address this problem, in this paper, we propose a Centralized Feature Pyramid (CFP) for object detection, which is based on a globally explicit centralized feature regulation. Specifically, we first propose a spatial explicit visual center scheme, where a lightweight MLP is used to capture the globally long-range dependencies and a parallel learnable visual center mechanism is used to capture the local corner regions of the input images. Based on this, we then propose a globally centralized regulation for the commonly-used feature pyramid in a top-down fashion, where the explicit visual center information obtained from the deepest intra-layer feature is used to regulate frontal shallow features. Compared to the existing feature pyramids, CFP not only has the ability to capture the global long-range dependencies, but also efficiently obtain an all-round yet discriminative feature representation. Experimental results on the challenging MS-COCO validate that our proposed CFP can achieve the consistent performance gains on the state-of-the-art YOLOv5 and YOLOX object detection baselines.
Few-shot fine-grained recognition (FS-FGR) aims to recognize novel fine-grained categories with the help of limited available samples. Undoubtedly, this task inherits the main challenges from both few-shot learning and fine-grained recognition. First, the lack of labeled samples makes the learned model easy to overfit. Second, it also suffers from high intra-class variance and low inter-class difference in the datasets. To address this challenging task, we propose a two-stage background suppression and foreground alignment framework, which is composed of a background activation suppression (BAS) module, a foreground object alignment (FOA) module, and a local to local (L2L) similarity metric. Specifically, the BAS is introduced to generate a foreground mask for localization to weaken background disturbance and enhance dominative foreground objects. What's more, considering the lack of labeled samples, we compute the pairwise similarity of feature maps using both the raw image and the refined image. The FOA then reconstructs the feature map of each support sample according to its correction to the query ones, which addresses the problem of misalignment between support-query image pairs. To enable the proposed method to have the ability to capture subtle differences in confused samples, we present a novel L2L similarity metric to further measure the local similarity between a pair of aligned spatial features in the embedding space. Extensive experiments conducted on multiple popular fine-grained benchmarks demonstrate that our method outperforms the existing state-of-the-art by a large margin.