Handling complicated text images (e.g., irregular structures, low resolution, heavy occlusion, and even illumination), existing supervised text recognition methods are data-hungry. Although these methods employ large-scale synthetic text images to reduce the dependence on annotated real images, the domain gap limits the recognition performance. Therefore, exploring the robust text feature representation on unlabeled real images by self-supervised learning is a good solution. However, existing self-supervised text recognition methods only execute sequence-to-sequence representation learning by roughly splitting the visual features along the horizontal axis, which will damage the character structures. Besides, these sequential-level self-learning methods limit the availability of geometric-based data augmentation, as large-scale geometry augmentation leads to sequence-to-sequence inconsistency. To address the above-mentioned issues, we proposed a novel self-supervised character-to-character distillation method, CCD. Specifically, we delineate the character structures of unlabeled real images by designing a self-supervised character segmentation module, and further apply the segmentation results to build character-level representation learning. CCD differs from prior works in that we propose a character-level pretext task to learn more fine-grained feature representations. Besides, compared with the inflexible augmentations of sequence-to-sequence models, our work satisfies character-to-character representation consistency, across various transformations (e.g., geometry and colour), to generate robust text features in the representative space. Experiments demonstrate that CCD achieves state-of-the-art performance on publicly available text recognition benchmarks.
We present X-Decoder, a generalized decoding model that can predict pixel-level segmentation and language tokens seamlessly. X-Decodert takes as input two types of queries: (i) generic non-semantic queries and (ii) semantic queries induced from text inputs, to decode different pixel-level and token-level outputs in the same semantic space. With such a novel design, X-Decoder is the first work that provides a unified way to support all types of image segmentation and a variety of vision-language (VL) tasks. Further, our design enables seamless interactions across tasks at different granularities and brings mutual benefits by learning a common and rich pixel-level visual-semantic understanding space, without any pseudo-labeling. After pretraining on a mixed set of a limited amount of segmentation data and millions of image-text pairs, X-Decoder exhibits strong transferability to a wide range of downstream tasks in both zero-shot and finetuning settings. Notably, it achieves (1) state-of-the-art results on open-vocabulary segmentation and referring segmentation on eight datasets; (2) better or competitive finetuned performance to other generalist and specialist models on segmentation and VL tasks; and (3) flexibility for efficient finetuning and novel task composition (e.g., referring captioning and image editing). Code, demo, video, and visualization are available at https://x-decoder-vl.github.io.
For applications that require processing large amounts of text at inference time, Large Language Models (LLMs) are handicapped by their limited context windows, which are typically 2048 tokens. In-context learning, an emergent phenomenon in LLMs in sizes above a certain parameter threshold, constitutes one significant example because it can only leverage training examples that fit into the context window. Existing efforts to address the context window limitation involve training specialized architectures, which tend to be smaller than the sizes in which in-context learning manifests due to the memory footprint of processing long texts. We present Parallel Context Windows (PCW), a method that alleviates the context window restriction for any off-the-shelf LLM without further training. The key to the approach is to carve a long context into chunks (``windows'') that fit within the architecture, restrict the attention mechanism to apply only within each window, and re-use the positional embeddings among the windows. We test the PCW approach on in-context learning with models that range in size between 750 million and 178 billion parameters, and show substantial improvements for tasks with diverse input and output spaces. Our results motivate further investigation of Parallel Context Windows as a method for applying off-the-shelf LLMs in other settings that require long text sequences.
Diffusion models have shown great promise for image generation, beating GANs in terms of generation diversity, with comparable image quality. However, their application to 3D shapes has been limited to point or voxel representations that can in practice not accurately represent a 3D surface. We propose a diffusion model for neural implicit representations of 3D shapes that operates in the latent space of an auto-decoder. This allows us to generate diverse and high quality 3D surfaces. We additionally show that we can condition our model on images or text to enable image-to-3D generation and text-to-3D generation using CLIP embeddings. Furthermore, adding noise to the latent codes of existing shapes allows us to explore shape variations.
We designed and built a game called \textit{Immersive Text Game}, which allows the player to choose a story and a character, and interact with other characters in the story in an immersive manner of dialogues. The game is based on several latest models, including text generation language model, information extraction model, commonsense reasoning model, and psychology evaluation model. In the past, similar text games usually let players choose from limited actions instead of answering on their own, and not every time what characters said are determined by the player. Through the combination of these models and elaborate game mechanics and modes, the player will find some novel experiences as driven through the storyline.
In this paper, we design and train a Generative Image-to-text Transformer, GIT, to unify vision-language tasks such as image/video captioning and question answering. While generative models provide a consistent network architecture between pre-training and fine-tuning, existing work typically contains complex structures (uni/multi-modal encoder/decoder) and depends on external modules such as object detectors/taggers and optical character recognition (OCR). In GIT, we simplify the architecture as one image encoder and one text decoder under a single language modeling task. We also scale up the pre-training data and the model size to boost the model performance. Without bells and whistles, our GIT establishes new state of the arts on 12 challenging benchmarks with a large margin. For instance, our model surpasses the human performance for the first time on TextCaps (138.2 vs. 125.5 in CIDEr). Furthermore, we present a new scheme of generation-based image classification and scene text recognition, achieving decent performance on standard benchmarks.
Interactive Machine Learning (IML) shall enable intelligent systems to interactively learn from their end-users, and is quickly becoming more and more important. Although it puts the human in the loop, interactions are mostly performed via mutual explanations that miss contextual information. Furthermore, current model-agnostic IML strategies like CAIPI are limited to 'destructive' feedback, meaning they solely allow an expert to prevent a learner from using irrelevant features. In this work, we propose a novel interaction framework called Semantic Interactive Learning for the text domain. We frame the problem of incorporating constructive and contextual feedback into the learner as a task to find an architecture that (a) enables more semantic alignment between humans and machines and (b) at the same time helps to maintain statistical characteristics of the input domain when generating user-defined counterexamples based on meaningful corrections. Therefore, we introduce a technique called SemanticPush that is effective for translating conceptual corrections of humans to non-extrapolating training examples such that the learner's reasoning is pushed towards the desired behavior. In several experiments, we show that our method clearly outperforms CAIPI, a state of the art IML strategy, in terms of Predictive Performance as well as Local Explanation Quality in downstream multi-class classification tasks.
Generating natural language statements to convey information from tabular data (i.e., Table-to-text) is a process with one input and a variety of valid outputs. This characteristic underscores the abilities to control the generation and produce a diverse set of outputs as two key assets. Thus, we propose a diversity enhancing scheme that builds upon an inherent property of the statements, namely, their logic-types, by using a type-controlled Table-to-text generation model. Employing automatic and manual tests, we prove its twofold advantage: users can effectively tune the generated statement type, and, by sampling different types, can obtain a diverse set of statements for a given table.
When generating text from probabilistic models, the chosen decoding strategy has a profound effect on the resulting text. Yet the properties elicited by various decoding strategies do not always transfer across natural language generation tasks. For example, while mode-seeking methods like beam search perform remarkably well for machine translation, they have been observed to lead to incoherent and repetitive text in story generation. Despite such observations, the effectiveness of decoding strategies is often assessed with respect to only a single task. This work -- in contrast -- provides a comprehensive analysis of the interaction between language generation tasks and decoding strategies. Specifically, we measure changes in attributes of generated text as a function of both decoding strategy and task using human and automatic evaluation. Our results reveal both previously-observed and surprising findings. For example, the nature of the diversity-quality trade-off in language generation is very task-specific; the length bias often attributed to beam search is not constant across tasks.
Segmentation-based methods have achieved great success for arbitrary shape text detection. However, separating neighboring text instances is still one of the most challenging problems due to the complexity of texts in scene images. In this paper, we propose an innovative Kernel Proposal Network (dubbed KPN) for arbitrary shape text detection. The proposed KPN can separate neighboring text instances by classifying different texts into instance-independent feature maps, meanwhile avoiding the complex aggregation process existing in segmentation-based arbitrary shape text detection methods. To be concrete, our KPN will predict a Gaussian center map for each text image, which will be used to extract a series of candidate kernel proposals (i.e., dynamic convolution kernel) from the embedding feature maps according to their corresponding keypoint positions. To enforce the independence between kernel proposals, we propose a novel orthogonal learning loss (OLL) via orthogonal constraints. Specifically, our kernel proposals contain important self-information learned by network and location information by position embedding. Finally, kernel proposals will individually convolve all embedding feature maps for generating individual embedded maps of text instances. In this way, our KPN can effectively separate neighboring text instances and improve the robustness against unclear boundaries. To our knowledge, our work is the first to introduce the dynamic convolution kernel strategy to efficiently and effectively tackle the adhesion problem of neighboring text instances in text detection. Experimental results on challenging datasets verify the impressive performance and efficiency of our method. The code and model are available at https://github.com/GXYM/KPN.