Vision transformers have gained popularity recently, leading to the development of new vision backbones with improved features and consistent performance gains. However, these advancements are not solely attributable to novel feature transformation designs; certain benefits also arise from advanced network-level and block-level architectures. This paper aims to identify the real gains of popular convolution and attention operators through a detailed study. We find that the key difference among these feature transformation modules, such as attention or convolution, lies in their spatial feature aggregation approach, known as the "spatial token mixer" (STM). To facilitate an impartial comparison, we introduce a unified architecture to neutralize the impact of divergent network-level and block-level designs. Subsequently, various STMs are integrated into this unified framework for comprehensive comparative analysis. Our experiments on various tasks and an analysis of inductive bias show a significant performance boost due to advanced network-level and block-level designs, but performance differences persist among different STMs. Our detailed analysis also reveals various findings about different STMs, including effective receptive fields, invariance, and adversarial robustness tests.

Video depth estimation aims to infer temporally consistent depth. One approach is to finetune a single-image model on each video with geometry constraints, which proves inefficient and lacks robustness. An alternative is learning to enforce consistency from data, which requires well-designed models and sufficient video depth data. To address both challenges, we introduce NVDS that stabilizes inconsistent depth estimated by various single-image models in a plug-and-play manner. We also elaborate a large-scale Video Depth in the Wild (VDW) dataset, which contains 14,203 videos with over two million frames, making it the largest natural-scene video depth dataset. Additionally, a bidirectional inference strategy is designed to improve consistency by adaptively fusing forward and backward predictions. We instantiate a model family ranging from small to large scales for different applications. The method is evaluated on VDW dataset and three public benchmarks. To further prove the versatility, we extend NVDS to video semantic segmentation and several downstream applications like bokeh rendering, novel view synthesis, and 3D reconstruction. Experimental results show that our method achieves significant improvements in consistency, accuracy, and efficiency. Our work serves as a solid baseline and data foundation for learning-based video depth estimation.

Generalizable NeRF aims to synthesize novel views for unseen scenes. Common practices involve constructing variance-based cost volumes for geometry reconstruction and encoding 3D descriptors for decoding novel views. However, existing methods show limited generalization ability in challenging conditions due to inaccurate geometry, sub-optimal descriptors, and decoding strategies. We address these issues point by point. First, we find the variance-based cost volume exhibits failure patterns as the features of pixels corresponding to the same point can be inconsistent across different views due to occlusions or reflections. We introduce an Adaptive Cost Aggregation (ACA) approach to amplify the contribution of consistent pixel pairs and suppress inconsistent ones. Un-like previous methods that solely fuse 2D features into descriptors, our approach introduces a Spatial-View Ag-gregator (SVA) to incorporate 3D context into descriptors through spatial and inter-view interaction. When decoding the descriptors, we observe the two existing decoding strategies excel in different areas, which are complementary. A Consistency-Aware Fusion (CAF) strategy is proposed to leverage the advantages of both. We incorporate the above ACA, SVA, and CAF into a coarse-to-fine framework, termed Geometry-aware Reconstruction and Fusion-refined Rendering (GeFu). GeFu attains state-of-the-art performance across multiple datasets. Code is available at https://github.com/TQTQliu/GeFu.

We present the All-Seeing (AS) project: a large-scale dataset and model for recognizing and understanding everything in the open world. Using a scalable data engine that incorporates human feedback and efficient models in the loop, we create a new dataset (AS-1B) with over 1.2 billion regions annotated with semantic tags, question-answering pairs, and detailed captions. It covers a wide range of 3.5 million common and rare concepts in the real world and has 132.2 billion tokens that describe the concepts and their attributes. Leveraging this new dataset, we develop the All-Seeing model (ASM), a unified framework for panoptic visual recognition and understanding. The model is trained with open-ended language prompts and locations, which allows it to generalize to various vision and language tasks with remarkable zero-shot performance, including both region- and image-level retrieval, region recognition, captioning, and question-answering. We hope that this project can serve as a foundation for vision-language artificial general intelligence research. Code is available at https://github.com/OpenGVLab/all-seeing.

Bokeh is a wide-aperture optical effect that creates aesthetic blurring in photography. However, achieving this effect typically demands expensive professional equipment and expertise. To make such cinematic techniques more accessible, bokeh rendering aims to generate the desired bokeh effects from all-in-focus inputs captured by smartphones. Previous efforts in bokeh rendering primarily focus on static images. However, when extended to video inputs, these methods exhibit flicker and artifacts due to a lack of temporal consistency modeling. Meanwhile, they cannot utilize information like occluded objects from adjacent frames, which are necessary for bokeh rendering. Moreover, the difficulties of capturing all-in-focus and bokeh video pairs result in a shortage of data for training video bokeh models. To tackle these challenges, we propose the Video Bokeh Renderer (VBR), the model designed specifically for video bokeh rendering.VBR leverages implicit feature space alignment and aggregation to model temporal consistency and exploit complementary information from adjacent frames. On the data front, we introduce the first Synthetic Video Bokeh (SVB) dataset, synthesizing authentic bokeh effects using ray-tracing techniques. Furthermore, to improve the robustness of the model to inaccurate disparity maps, we employ a set of augmentation strategies to simulate corrupted disparity inputs during training. Experimental results on both synthetic and real-world data demonstrate the effectiveness of our method.

Category-agnostic pose estimation (CAPE) aims to predict keypoints for arbitrary categories given support images with keypoint annotations. Existing approaches match the keypoints across the image for localization. However, such a one-stage matching paradigm shows inferior accuracy: the prediction heavily relies on the matching results, which can be noisy due to the open set nature in CAPE. For example, two mirror-symmetric keypoints (e.g., left and right eyes) in the query image can both trigger high similarity on certain support keypoints (eyes), which leads to duplicated or opposite predictions. To calibrate the inaccurate matching results, we introduce a two-stage framework, where matched keypoints from the first stage are viewed as similarity-aware position proposals. Then, the model learns to fetch relevant features to correct the initial proposals in the second stage. We instantiate the framework with a transformer model tailored for CAPE. The transformer encoder incorporates specific designs to improve the representation and similarity modeling in the first matching stage. In the second stage, similarity-aware proposals are packed as queries in the decoder for refinement via cross-attention. Our method surpasses the previous best approach by large margins on CAPE benchmark MP-100 on both accuracy and efficiency. Code available at github.com/flyinglynx/CapeFormer

Video depth estimation aims to infer temporally consistent depth. Some methods achieve temporal consistency by finetuning a single-image depth model during test time using geometry and re-projection constraints, which is inefficient and not robust. An alternative approach is to learn how to enforce temporal consistency from data, but this requires well-designed models and sufficient video depth data. To address these challenges, we propose a plug-and-play framework called Neural Video Depth Stabilizer (NVDS) that stabilizes inconsistent depth estimations and can be applied to different single-image depth models without extra effort. We also introduce a large-scale dataset, Video Depth in the Wild (VDW), which consists of 14,203 videos with over two million frames, making it the largest natural-scene video depth dataset to our knowledge. We evaluate our method on the VDW dataset as well as two public benchmarks and demonstrate significant improvements in consistency, accuracy, and efficiency compared to previous approaches. Our work serves as a solid baseline and provides a data foundation for learning-based video depth models. We will release our dataset and code for future research.

Learning-based multi-view stereo (MVS) method heavily relies on feature matching, which requires distinctive and descriptive representations. An effective solution is to apply non-local feature aggregation, e.g., Transformer. Albeit useful, these techniques introduce heavy computation overheads for MVS. Each pixel densely attends to the whole image. In contrast, we propose to constrain nonlocal feature augmentation within a pair of lines: each point only attends the corresponding pair of epipolar lines. Our idea takes inspiration from the classic epipolar geometry, which shows that one point with different depth hypotheses will be projected to the epipolar line on the other view. This constraint reduces the 2D search space into the epipolar line in stereo matching. Similarly, this suggests that the matching of MVS is to distinguish a series of points lying on the same line. Inspired by this point-toline search, we devise a line-to-point non-local augmentation strategy. We first devise an optimized searching algorithm to split the 2D feature maps into epipolar line pairs. Then, an Epipolar Transformer (ET) performs non-local feature augmentation among epipolar line pairs. We incorporate the ET into a learning-based MVS baseline, named ET-MVSNet. ET-MVSNet achieves state-of-the-art reconstruction performance on both the DTU and Tanks-and-Temples benchmark with high efficiency. Code is available at https://github.com/TQTQliu/ET-MVSNet.

Human Mesh Reconstruction (HMR) aims to recover 3D human poses and shapes from a single image. Existing parameterized HMR approaches follow the "representation-to-reasoning'' paradigm to predict human body and pose parameters. This paradigm typically involves intermediate representation and complex pipeline, where potential side effects may occur that could hinder performance. In contrast, query-based non-parameterized methods directly output 3D joints and mesh vertices, but they rely on excessive queries for prediction, leading to low efficiency and robustness. In this work, we propose a simple query-based framework, dubbed SimHMR, for parameterized human mesh reconstruction. This framework streamlines the prediction process by using a few parameterized queries, which effectively removes the need for hand-crafted intermediate representation and reasoning pipeline. Different from query-based non-parameterized HMR that uses excessive coordinate queries, SimHMR only requires a few semantic queries, which physically correspond to pose, shape, and camera. The use of semantic queries significantly improves the efficiency and robustness in extreme scenarios, e.g., occlusions. Without bells and whistles, øurs achieves state-of-the-art performance on 3DPW and Human3.6M benchmarks, and surpasses existing methods on challenging 3DPW-OCC. Code available at https://github.com/inso-13/SimHMR github.com/inso-13/SimHMR

Class-agnostic counting (CAC) aims to count all instances in a query image given few exemplars. A standard pipeline is to extract visual features from exemplars and match them with query images to infer object counts. Two essential components in this pipeline are feature representation and similarity metric. Existing methods either adopt a pretrained network to represent features or learn a new one, while applying a naive similarity metric with fixed inner product. We find this paradigm leads to noisy similarity matching and hence harms counting performance. In this work, we propose a similarity-aware CAC framework that jointly learns representation and similarity metric. We first instantiate our framework with a naive baseline called Bilinear Matching Network (BMNet), whose key component is a learnable bilinear similarity metric. To further embody the core of our framework, we extend BMNet to BMNet+ that models similarity from three aspects: 1) representing the instances via their self-similarity to enhance feature robustness against intra-class variations; 2) comparing the similarity dynamically to focus on the key patterns of each exemplar; 3) learning from a supervision signal to impose explicit constraints on matching results. Extensive experiments on a recent CAC dataset FSC147 show that our models significantly outperform state-of-the-art CAC approaches. In addition, we also validate the cross-dataset generality of BMNet and BMNet+ on a car counting dataset CARPK. Code is at tiny.one/BMNet

Generative adversarial networks,(GANs) have been trained to be professional artists able to create stunning artworks such as face generation and image style transfer. In this paper, we focus on a realistic business scenario: automated generation of customizable icons given desired mobile applications and theme styles. We first introduce a theme-application icon dataset, termed AppIcon, where each icon has two orthogonal theme and app labels. By investigating a strong baseline StyleGAN2, we observe mode collapse caused by the entanglement of the orthogonal labels. To solve this challenge, we propose IconGAN composed of a conditional generator and dual discriminators with orthogonal augmentations, and a contrastive feature disentanglement strategy is further designed to regularize the feature space of the two discriminators. Compared with other approaches, IconGAN indicates a superior advantage on the AppIcon benchmark. Further analysis also justifies the effectiveness of disentangling app and theme representations. Our project will be released at: https://github.com/architect-road/IconGAN.