ECCV 2022 Papers

"Text-Video retrieval is a task of great practical value and has received increasing attention, among which learning spatial-temporal video representation is one of the research hotspots. The video encoders in the state-of-the-art video retrieval models usually directly adopt the pre-trained vision backbones with the network structure fixed, they therefore can not be further improved to produce the fine-grained spatial-temporal video representation. In this paper, we propose Token Shift and Selection Network (TS2-Net), a novel token shift and selection transformer architecture, which dynamically adjusts the token sequence and selects informative tokens in both temporal and spatial dimensions from input video samples. The token shift module temporally shifts the whole token features back-and-forth across adjacent frames, to preserve the complete token representation and capture subtle movements. Then the token selection module selects tokens that contribute most to local spatial semantics. Based on thorough experiments, the proposed TS2-Net achieves state-of-the-art performance on major text-video retrieval benchmarks, including new records on MSRVTT, VATEX, LSMDC, ActivityNet, and DiDeMo. Code is available at https://github.com/yuqi657/ts2_net."

"High-quality structured data with rich annotations are critical components in intelligent vehicle systems dealing with road scenes. However, data curation and annotation require intensive investments and yield low-diversity scenarios. The recently growing interest in synthetic data raises questions about the scope of improvement in such systems and the amount of manual work still required to produce high volumes and variations of simulated data. This work proposes a synthetic data generation pipeline that utilizes existing datasets, like nuScenes, to address the difficulties and domain-gaps present in simulated datasets. We show that using annotations and visual cues from existing datasets, we can facilitate automated multi-modal data generation, mimicking real scene properties with high-fidelity, along with mechanisms to diversify samples in a physically meaningful way. We demonstrate improvements in mIoU metrics by presenting qualitative and quantitative experiments with real and synthetic data for semantic segmentation on the Cityscapes and KITTI-STEP datasets. All relevant code and data is released on github."

"Evaluating image generation models such as generative adversarial networks (GANs) is a challenging problem. A common approach is to compare the distributions of the set of ground truth images and the set of generated test images. The Frechet Inception distance is one of the most widely used metrics for evaluation of GANs, which assumes that the features from a trained Inception model for a set of images follow a normal distribution. In this paper, we argue that this is an over-simplified assumption, which may lead to unreliable evaluation results, and more accurate density estimation can be achieved using a truncated generalized normal distribution. Based on this, we propose a novel metric for accurate evaluation of GANs, named TREND (TRuncated gEneralized Normal Density estimation of inception embeddings). We demonstrate that our approach significantly reduces errors of density estimation, which consequently eliminates the risk of faulty evaluation results. Furthermore, we show that the proposed metric significantly improves robustness of evaluation results against variation of the number of image samples."

"We present TOCH, a method for refining incorrect 3D hand-object interaction sequences using a data prior. Existing hand trackers, especially those that rely on very few cameras, often produce visually unrealistic results with hand-object intersection or missing contacts. Although correcting such errors requires reasoning about temporal aspects of interaction, most previous works focus on static grasps and contacts. The core of our method are TOCH fields, a novel spatio-temporal representation for modeling correspondences between hands and objects during interaction. TOCH fields are a point-wise, object-centric representation, which encode the hand position relative to the object. Leveraging this novel representation, we learn a latent manifold of plausible TOCH fields with a temporal denoising auto-encoder. Experiments demonstrate that TOCH outperforms state-of-the-art 3D hand-object interaction models, which are limited to static grasps and contacts. More importantly, our method produces smooth interactions even before and after contact. Using a single trained TOCH model, we quantitatively and qualitatively demonstrate its usefulness for correcting erroneous sequences from off-the-shelf RGB/RGB-D hand-object reconstruction methods and transferring grasps across objects."

"Many basic indoor activities such as eating or writing are always conducted upon different tabletops (e.g., coffee tables, writing desks). It is indispensable to understanding tabletop scenes in 3D indoor scene parsing applications. Unfortunately, it is hard to meet this demand by directly deploying data-driven algorithms, since 3D tabletop scenes are rarely available in current datasets. To remedy this defect, we introduce TO-Scene, a large-scale dataset focusing on tabletop scenes, which contains 20,740 scenes with three variants. To acquire the data, we design an efficient and scalable framework, where a crowdsourcing UI is developed to transfer CAD objects from ModelNet and ShapeNet onto tables from ScanNet, then the output tabletop scenes are simulated into real scans and annotated automatically. Further, a tabletop-aware learning strategy is proposed for better perceiving the small-sized tabletop instances. Notably, we also provide a real scanned test set TO-Real to verify the practical value of TO-Scene. Experiments show that the algorithms trained on TO-Scene indeed work on the realistic test data, and our proposed tabletop-aware learning strategy greatly improves the state-of-the-art results on both 3D semantic segmentation and object detection tasks. Dataset and code are available at https://github.com/GAP-LAB-CUHK-SZ/TO-Scene."

"Inspired by the strong ties between vision and language, the two intimate human sensing and communication modalities, our paper aims to explore the generation of 3D human full-body motions from texts, as well as its reciprocal task, shorthanded for text2motion and motion2text, respectively. To tackle the existing challenges, especially to enable the generation of multiple distinct motions from the same text, and to avoid the undesirable production of trivial motionless pose sequences, we propose the use of motion token, a discrete and compact motion representation, where motions and texts could then be considered on one level playing ground, as the motion and text tokens. Moreover, our motion2text module is integrated into the inverse alignment process of our text2motion training pipeline, where a significant deviation of synthesized text (text2motion-2text) from the input text would be penalized by a large training loss; empirically this is shown to achieve improved performance. Finally, the mappings in-between the two modalities of motions and texts are facilitated by adapting the neural model for machine translation (NMT) to our context. Autoregressive modeling on the underlying distribution of discrete motion tokens further enables the production of non-deterministic motions from texts. Overall our approach is flexible, and could be used for both text2motion and motion2text tasks. Empirical evaluations on two benchmark datasets demonstrate the superior performance of our approach over a variety of state-of-the-art methods on both tasks."

"YouTube users looking for instructions for a specific task may spend a long time browsing content trying to find the right video that matches their needs. Creating a visual summary (abridged version of a video) provides viewers with a quick overview and massively reduces search time. In this work, we focus on summarizing instructional videos, an under-explored area of video summarization. In comparison to generic videos, instructional videos can be parsed into semantically meaningful segments that correspond to important steps of the demonstrated task. Existing video summarization datasets rely on manual frame-level annotations, making them subjective and limited in size. To overcome this, we first automatically generate pseudo summaries for a corpus of instructional videos by exploiting two key assumptions: (i) relevant steps are likely to appear in multiple videos of the same task (Task Relevance), and (ii) they are more likely to be described by the demonstrator verbally (Cross-Modal Saliency). We propose an instructional video summarization network that combines a context-aware temporal video encoder and a segment scoring transformer. Using pseudo summaries as weak supervision, our network constructs a visual summary for an instructional video given only video and transcribed speech. To evaluate our model, we collect a high-quality test set, WikiHow Summaries, by scraping WikiHow articles that contain video demonstrations and visual depictions of steps allowing us to obtain the ground-truth summaries. We outperform several baselines and a state-of-the-art video summarization model on this new benchmark."

"In this paper, we conduct a study on the state-of-the-art methods for text-to-image synthesis and propose a framework to evaluate these methods. We consider syntheses where an image contains a single or multiple objects. Our study outlines several issues in the current evaluation pipeline: (i) for image quality assessment, a commonly used metric, e.g., Inception Score (IS), is often either miscalibrated for the single-object case or misused for the multi-object case; (ii) for text relevance and object accuracy assessment, there is an overfitting phenomenon in the existing R-precision (RP) and Semantic Object Accuracy (SOA) metrics, respectively; (iii) for multi-object case, many vital factors for evaluation, e.g., object fidelity, positional alignment, counting alignment, are largely dismissed; (iv) the ranking of the methods based on current metrics is highly inconsistent with real images. To overcome these issues, we propose a combined set of existing and new metrics to systematically evaluate the methods. For existing metrics, we offer an improved version of IS named IS* by using temperature scaling to calibrate the confidence of the classifier used by IS; we also propose a solution to mitigate the overfitting issues of RP and SOA. For new metrics, we develop counting alignment, positional alignment, object-centric IS, and object-centric FID metrics for evaluating the multi-object case. We show that benchmarking with our bag of metrics results in a highly consistent ranking among existing methods that is well-aligned with human evaluation. As a by-product, we create AttnGAN++, a simple but strong baseline for the benchmark by stabilizing the training of AttnGAN using spectral normalization. We also release our toolbox, so-called TISE, for advocating fair and consistent evaluation of text-to-image models."

"In computer vision, human pose synthesis and transfer deal with probabilistic image generation of a person in a previously unseen pose from an already available observation of that person. Though researchers have recently proposed several methods to achieve this task, most of these techniques derive the target pose directly from the desired target image on a specific dataset, making the underlying process challenging to apply in real-world scenarios as the generation of the target image is the actual aim. In this paper, we first present the shortcomings of current pose transfer algorithms and then propose a novel text-based pose transfer technique to address those issues. We divide the problem into three independent stages: (a) text to pose representation, (b) pose refinement, and (c) pose rendering. To the best of our knowledge, this is one of the first attempts to develop a text-based pose transfer framework where we also introduce a new dataset DF-PASS, by adding descriptive pose annotations for the images of the DeepFashion dataset. The proposed method generates promising results with significant qualitative and quantitative scores in our experiments."

"We introduce TIDEE, an embodied agent that tidies up a disordered scene based on learned commonsense object placement and room arrangement priors. TIDEE explores a home environment, detects objects that are out of their natural place, infers plausible object contexts for them, localizes such contexts in the current scene, and repositions the objects. Commonsense priors are encoded in three modules: i) visuo-semantic detectors that detect out-of-place objects, ii) an associative neural graph memory of objects and spatial relations that proposes plausible semantic receptacles and surfaces for object repositions, and iii) a visual search network that guides the agent’s exploration for efficiently localizing the receptacle-of-interest in the current scene to reposition the object. We test TIDEE on tidying up disorganized scenes in the AI2THOR simulation environment. TIDEE carries out the task directly from pixel and raw depth input without ever having observed the same room beforehand, relying only on priors learned from a separate set of training houses. Human evaluations on the resulting room reorganizations show TIDEE outperforms ablative versions of the model that do not use one or more of the commonsense priors. On a related room rearrangement benchmark that allows the agent to view the goal state prior to rearrangement, a simplified version of our model significantly outperforms a top-performing method by a large margin. Code and data are available at the project website: https://tidee-agent.github.io/."

"We address the problem of generating diverse 3D human motions from textual descriptions. This challenging task requires joint modeling of both modalities: understanding and extracting useful human-centric information from the text, and then generating plausible and realistic sequences of human poses. In contrast to most previous work which focuses on generating a single, deterministic, motion from a textual description, we design a variational approach that can produce multiple diverse human motions. We propose TEMOS, a text-conditioned generative model leveraging variational autoencoder (VAE) training with human motion data, in combination with a text encoder that produces distribution parameters compatible with the VAE latent space. We show the TEMOS framework can produce both skeleton-based animations as in prior work, as well more expressive SMPL body motions. We evaluate our approach on the KIT Motion-Language benchmark and, despite being relatively straightforward, demonstrate significant improvements over the state of the art. Code and models are available on our webpage."

"Deep video models, for example, 3D CNNs or video transformers, have achieved promising performance on sparse video tasks, i.e., predicting one result per video. However, challenges arise when adapting existing deep video models to dense video tasks, i.e., predicting one result per frame. Specifically, these models are expensive for deployment, less effective when handling redundant frames and difficult to capture long-range temporal correlations. To overcome these issues, we propose a Temporal Dilated Video Transformer (TDViT) that consists of carefully-designed temporal dilated transformer blocks (TDTB). TDTB can efficiently extract spatiotemporal representations and effectively alleviate the negative effect of temporal redundancy. Furthermore, by using hierarchical TDTBs, our approach obtains an exponentially expanded temporal receptive field and therefore can model long-range dynamics. Extensive experiments are conducted on two different dense video benchmarks, i.e., ImageNet VID for video object detection and YouTube VIS for video instance segmentation. Excellent experimental results demonstrate the superior efficiency, effectiveness, and compatibility of our method. The code is available at https://github.com/guanxiongsun/TDViT."

"Attention modules for Convolutional Neural Networks (CNNs) are an effective method to enhance performance on multiple computer-vision tasks. While existing methods appropriately model channel-, spatial- and self-attention, they primarily operate in a feedforward bottom-up manner. Consequently, the attention mechanism strongly depends on the local information of a single input feature map and does not incorporate relatively semantically-richer contextual information available at higher layers that can specify “what and where to look” in lower-level feature maps through top-down information flow. Accordingly, in this work, we propose a lightweight top-down attention module (TDAM) that iteratively generates a “visual searchlight” to perform channel and spatial modulation of its inputs and outputs more contextually-relevant feature maps at each computation step. Our experiments indicate that TDAM enhances the performance of CNNs across multiple object-recognition benchmarks and outperforms prominent attention modules while being more parameter and memory efficient. Further, TDAM-based models learn to “shift attention” by localizing individual objects or features at each computation step without any explicit supervision resulting in a 5% improvement for ResNet50 on weakly-supervised object localization. Source code and models are publicly available at: https://github.com/shantanuj/TDAM_Top_down_attention_module"

"Graph-based approaches have been becoming increasingly popular in road network extraction, in addition to segmentation-based methods. Road networks are represented as graph structures, being able to explicitly define the topology structures and avoid the ambiguity of segmentation masks, such as between a real junction area and multiple separate roads in different heights. In contrast to the bottom-up graph-based approaches, which rely on orientation information, we propose a novel top-down approach to generate road network graphs with a holistic model, namely TD-Road. We decompose road extraction as two subtasks: key point prediction and connectedness prediction. We directly apply graph structures (i.e., locations of node and connections between them) as training supervisions for neural networks and generate road graph outputs in inference, instead of learning some intermediate properties of a graph structure (e.g., orientations or distances for the next move). Our network integrates a relation inference module with key point prediction, to capture connections between neighboring points and outputs the final road graphs with no post-processing steps required. Extensive experiments are conducted on challenging datasets, including City-Scale and SpaceNet to show the effectiveness and simplicity of our method, that the proposed method achieves remarkable results compared with previous state-of-the-art methods."

"Coordinate-based volumetric representations have the potential to generate photo-realistic virtual avatars from images. However, virtual avatars need to be controllable and be rendered in novel poses that may not have been observed. Traditional techniques, such as LBS, provide such a controlling function; yet it usually requires a hand-designed body template, 3D scan data, and surface-based appearance models. On the other hand, neural representations have been shown to be powerful in representing visual details, but are under-explored in dynamic and articulated settings. In this paper, we propose TAVA, a method to create Template-free Animatable Volumetric Actors, based on neural representations. We rely solely on multi-view data and a tracked skeleton to create a volumetric model of an actor, which can be animated at test time given novel poses. Since TAVA does not require a body template, it is applicable to humans as well as other creatures such as animals. Furthermore, TAVA is designed such that it can recover accurate dense correspondences, making it amenable to content-creation and editing tasks. Through extensive experiments, we demonstrate that he proposed method generalizes well to novel poses as well as unseen views and showcase basic editing capabilities. The code is available at https://github.com/facebookresearch/tava"

"Learning a generalized prior for natural image restoration is an important yet challenging task. Early methods mostly involved handcrafted priors including normalized sparsity, â„“0 gradients, dark channel priors, etc.. Recently, deep neural networks have been used to learn various image priors but do not guarantee to generalize. In this paper, we propose a novel approach that embeds a task-agnostic prior into a transformer. Our approach, named Task-Agnostic Prior Embedding (TAPE), consists of two stages, namely, task-agnostic pre-training and task-specific fine-tuning, where the first stage embeds prior knowledge about natural images into the transformer and the second stage extracts the knowledge to assist downstream image restoration. Experiments on various types of degradation validate the effectiveness of TAPE. The image restoration performance in terms of PSNR is improved by as much as 1.45 dB and even outperforms task-specific algorithms. More importantly, TAPE shows the ability of disentangling generalized image priors from degraded images, which enjoys favorable transfer ability to unknown downstream tasks."

"Most modern approaches in temporal action localization divide this problem into two parts: (i) short-term feature extraction and (ii) long-range temporal boundary localization. Due to the high GPU memory cost caused by processing long untrimmed videos, many methods sacrifice the representational power of the short-term feature extractor by either freezing the backbone or using a small spatial video resolution. This issue becomes even worse with the recent video transformer models, many of which have quadratic memory complexity. To address these issues, we propose TALLFormer, a memory-efficient and end-to-end trainable Temporal Action Localization Transformer with Long-term memory. Our long-term memory mechanism eliminates the need for processing hundreds of redundant video frames during each training iteration, thus, significantly reducing the GPU memory consumption and training time. These efficiency savings allow us (i) to use a powerful video transformer feature extractor without freezing the backbone or reducing the spatial video resolution, while (ii) also maintaining long-range temporal boundary localization capability. With only RGB frames as input and no external action recognition classifier, TALLFormer outperforms previous state-of-the-arts by a large margin, achieving an average mAP of 59.1% on THUMOS14 and 35.6% on ActivityNet-1.3. The code is public available: https://github.com/klauscc/TALLFormer."

"Deep neural networks based object detectors have shown great success in a variety of domains like autonomous vehicles, biomedical imaging, etc. It is known that their success depends on a large amount of data from the domain of interest. While deep models often perform well in terms of overall accuracy, they often struggle in performance on rare yet critical data slices. For example, data slices like ""motorcycle at night"" or ""bicycle at night"" are often rare but very critical slices for self-driving applications and false negatives on such rare slices could result in ill-fated failures and accidents. Active learning (AL) is a well-known paradigm to incrementally and adaptively build training datasets with a human in the loop. However, current AL based acquisition functions are not well-equipped to tackle real-world datasets with rare slices, since they are based on uncertainty scores or global descriptors of the image. We propose TALISMAN, a novel framework for Targeted Active Learning or object detectIon with rare slices using Submodular MutuAl iNformation. Our method uses the submodular mutual information functions instantiated using features of the region of interest (RoI) to efficiently target and acquire data points with rare slices. We evaluate our framework on the standard PASCAL VOC07+12 and BDD100K, a real-world self-driving dataset. We observe that TALISMAN outperforms other methods by in terms of average precision on rare slices, and in terms of mAP."

"Face manipulation methods can be misused to affect an individual’s privacy or to spread disinformation. To this end, we introduce a novel data-driven approach that produces image-specific perturbations which are embedded in the original images. The key idea is that these protected images prevent face manipulation by causing the manipulation model to produce a predefined manipulation target (uniformly colored output image in our case) instead of the actual manipulation. In addition, we propose to leverage differentiable compression approximation, hence making generated perturbations robust to common image compression. In order to prevent against multiple manipulation methods simultaneously, we further propose a novel attention-based fusion of manipulation-specific perturbations. Compared to traditional adversarial attacks that optimize noise patterns for each image individually, our generalized model only needs a single forward pass, thus running orders of magnitude faster and allowing for easy integration in image processing stacks, even on resource-constrained devices like smartphones"