We present YORO - a multi-modal transformer encoder-only architecture for the Visual Grounding (VG) task. This task involves localizing, in an image, an object referred via natural language. Unlike the recent trend in the literature of using multi-stage approaches that sacrifice speed for accuracy, YORO seeks a better trade-off between speed an accuracy by embracing a single-stage design, without CNN backbone. YORO consumes natural language queries, image patches, and learnable detection tokens and predicts coordinates of the referred object, using a single transformer encoder. To assist the alignment between text and visual objects, a novel patch-text alignment loss is proposed. Extensive experiments are conducted on 5 different datasets with ablations on architecture design choices. YORO is shown to support real-time inference and outperform all approaches in this class (single-stage methods) by large margins. It is also the fastest VG model and achieves the best speed/accuracy trade-off in the literature.
Grouping and recognition are important components of visual scene understanding, e.g., for object detection and semantic segmentation. With end-to-end deep learning systems, grouping of image regions usually happens implicitly via top-down supervision from pixel-level recognition labels. Instead, in this paper, we propose to bring back the grouping mechanism into deep networks, which allows semantic segments to emerge automatically with only text supervision. We propose a hierarchical Grouping Vision Transformer (GroupViT), which goes beyond the regular grid structure representation and learns to group image regions into progressively larger arbitrary-shaped segments. We train GroupViT jointly with a text encoder on a large-scale image-text dataset via contrastive losses. With only text supervision and without any pixel-level annotations, GroupViT learns to group together semantic regions and successfully transfers to the task of semantic segmentation in a zero-shot manner, i.e., without any further fine-tuning. It achieves a zero-shot accuracy of 51.2% mIoU on the PASCAL VOC 2012 and 22.3% mIoU on PASCAL Context datasets, and performs competitively to state-of-the-art transfer-learning methods requiring greater levels of supervision. Project page is available at https://jerryxu.net/GroupViT.
Novel architectures have recently improved generative image synthesis leading to excellent visual quality in various tasks. Of particular note is the field of ``AI-Art'', which has seen unprecedented growth with the emergence of powerful multimodal models such as CLIP. By combining speech and image synthesis models, so-called ``prompt-engineering'' has become established, in which carefully selected and composed sentences are used to achieve a certain visual style in the synthesized image. In this note, we present an alternative approach based on retrieval-augmented diffusion models (RDMs). In RDMs, a set of nearest neighbors is retrieved from an external database during training for each training instance, and the diffusion model is conditioned on these informative samples. During inference (sampling), we replace the retrieval database with a more specialized database that contains, for example, only images of a particular visual style. This provides a novel way to prompt a general trained model after training and thereby specify a particular visual style. As shown by our experiments, this approach is superior to specifying the visual style within the text prompt. We open-source code and model weights at https://github.com/CompVis/latent-diffusion .
Generating motion in line with text has attracted increasing attention nowadays. However, open-vocabulary human motion generation still remains touchless and undergoes the lack of diverse labeled data. The good news is that, recent studies of large multi-model foundation models (e.g., CLIP) have demonstrated superior performance on few/zero-shot image-text alignment, largely reducing the need for manually labeled data. In this paper, we take advantage of CLIP for open-vocabulary 3D human motion generation in a zero-shot manner. Specifically, our model is composed of two stages, i.e., text2pose and pose2motion. For text2pose, to address the difficulty of optimization with direct supervision from CLIP, we propose to carve the versatile CLIP model into a slimmer but more specific model for aligning 3D poses and texts, via a novel pipeline distillation strategy. Optimizing with the distilled 3D pose-text model, we manage to concretize the text-pose knowledge of CLIP into a text2pose generator effectively and efficiently. As for pose2motion, drawing inspiration from the advanced language model, we pretrain a transformer-based motion model, which makes up for the lack of motion dynamics of CLIP. After that, by formulating the generated poses from the text2pose stage as prompts, the motion generator can generate motions referring to the poses in a controllable and flexible manner. Our method is validated against advanced baselines and obtains sharp improvements. The code will be released here.
Story visualization advances the traditional text-to-image generation by enabling multiple image generation based on a complete story. This task requires machines to 1) understand long text inputs and 2) produce a globally consistent image sequence that illustrates the contents of the story. A key challenge of consistent story visualization is to preserve characters that are essential in stories. To tackle the challenge, we propose to adapt a recent work that augments Vector-Quantized Variational Autoencoders (VQ-VAE) with a text-tovisual-token (transformer) architecture. Specifically, we modify the text-to-visual-token module with a two-stage framework: 1) character token planning model that predicts the visual tokens for characters only; 2) visual token completion model that generates the remaining visual token sequence, which is sent to VQ-VAE for finalizing image generations. To encourage characters to appear in the images, we further train the two-stage framework with a character-token alignment objective. Extensive experiments and evaluations demonstrate that the proposed method excels at preserving characters and can produce higher quality image sequences compared with the strong baselines. Codes can be found in https://github.com/sairin1202/VP-CSV
Event detection (ED) identifies and classifies event triggers from unstructured texts, serving as a fundamental task for information extraction. Despite the remarkable progress achieved in the past several years, most research efforts focus on detecting events from formal texts (e.g., news articles, Wikipedia documents, financial announcements). Moreover, the texts in each dataset are either from a single source or multiple yet relatively homogeneous sources. With massive amounts of user-generated text accumulating on the Web and inside enterprises, identifying meaningful events in these informal texts, usually from multiple heterogeneous sources, has become a problem of significant practical value. As a pioneering exploration that expands event detection to the scenarios involving informal and heterogeneous texts, we propose a new large-scale Chinese event detection dataset based on user reviews, text conversations, and phone conversations in a leading e-commerce platform for food service. We carefully investigate the proposed dataset's textual informality and multi-source heterogeneity characteristics by inspecting data samples quantitatively and qualitatively. Extensive experiments with state-of-the-art event detection methods verify the unique challenges posed by these characteristics, indicating that multi-source informal event detection remains an open problem and requires further efforts. Our benchmark and code are released at \url{https://github.com/myeclipse/MUSIED}.
Realizing when a model is right for a wrong reason is not trivial and requires a significant effort by model developers. In some cases, an input salience method, which highlights the most important parts of the input, may reveal problematic reasoning. But scrutinizing highlights over many data instances is tedious and often infeasible. Furthermore, analyzing examples in isolation does not reveal general patterns in the data or in the model's behavior.In this paper we aim to address these issues and go from understanding single examples to understanding entire datasets and models. The methodology we propose is based on aggregated salience maps. Using this methodology we address multiple distinct but common model developer needs by showing how problematic data and model behavior can be identified -- a necessary first step for improving the model.
Effective image retrieval with text feedback stands to impact a range of real-world applications, such as e-commerce. Given a source image and text feedback that describes the desired modifications to that image, the goal is to retrieve the target images that resemble the source yet satisfy the given modifications by composing a multi-modal (image-text) query. We propose a novel solution to this problem, Additive Attention Compositional Learning (AACL), that uses a multi-modal transformer-based architecture and effectively models the image-text contexts. Specifically, we propose a novel image-text composition module based on additive attention that can be seamlessly plugged into deep neural networks. We also introduce a new challenging benchmark derived from the Shopping100k dataset. AACL is evaluated on three large-scale datasets (FashionIQ, Fashion200k, and Shopping100k), each with strong baselines. Extensive experiments show that AACL achieves new state-of-the-art results on all three datasets.
We summarize our TRECVID 2022 Ad-hoc Video Search (AVS) experiments. Our solution is built with two new techniques, namely Lightweight Attentional Feature Fusion (LAFF) for combining diverse visual / textual features and Bidirectional Negation Learning (BNL) for addressing queries that contain negation cues. In particular, LAFF performs feature fusion at both early and late stages and at both text and video ends to exploit diverse (off-the-shelf) features. Compared to multi-head self attention, LAFF is much more compact yet more effective. Its attentional weights can also be used for selecting fewer features, with the retrieval performance mostly preserved. BNL trains a negation-aware video retrieval model by minimizing a bidirectionally constrained loss per triplet, where a triplet consists of a given training video, its original description and a partially negated description. For video feature extraction, we use pre-trained CLIP, BLIP, BEiT, ResNeXt-101 and irCSN. As for text features, we adopt bag-of-words, word2vec, CLIP and BLIP. Our training data consists of MSR-VTT, TGIF and VATEX that were used in our previous participation. In addition, we automatically caption the V3C1 collection for pre-training. The 2022 edition of the TRECVID benchmark has again been a fruitful participation for the RUCMM team. Our best run, with an infAP of 0.262, is ranked at the second place teamwise.
Text to speech (TTS) has made rapid progress in both academia and industry in recent years. Some questions naturally arise that whether a TTS system can achieve human-level quality, how to define/judge that quality and how to achieve it. In this paper, we answer these questions by first defining the human-level quality based on the statistical significance of subjective measure and introducing appropriate guidelines to judge it, and then developing a TTS system called NaturalSpeech that achieves human-level quality on a benchmark dataset. Specifically, we leverage a variational autoencoder (VAE) for end-to-end text to waveform generation, with several key modules to enhance the capacity of the prior from text and reduce the complexity of the posterior from speech, including phoneme pre-training, differentiable duration modeling, bidirectional prior/posterior modeling, and a memory mechanism in VAE. Experiment evaluations on popular LJSpeech dataset show that our proposed NaturalSpeech achieves -0.01 CMOS (comparative mean opinion score) to human recordings at the sentence level, with Wilcoxon signed rank test at p-level p >> 0.05, which demonstrates no statistically significant difference from human recordings for the first time on this dataset.