Search

We introduce HyperCLOVA X, a family of large language models (LLMs) tailored to the Korean language and culture, along with competitive capabilities in English, math, and coding. HyperCLOVA X was trained on a balanced mix of Korean, English, and code data, followed by instruction-tuning with high-quality human-annotated datasets while abiding by strict safety guidelines reflecting our commitment to responsible AI. The model is evaluated across various benchmarks, including comprehensive reasoning, knowledge, commonsense, factuality, coding, math, chatting, instruction-following, and harmlessness, in both Korean and English. HyperCLOVA X exhibits strong reasoning capabilities in Korean backed by a deep understanding of the language and cultural nuances. Further analysis of the inherent bilingual nature and its extension to multilingualism highlights the model's cross-lingual proficiency and strong generalization ability to untargeted languages, including machine translation between several language pairs and cross-lingual inference tasks. We believe that HyperCLOVA X can provide helpful guidance for regions or countries in developing their sovereign LLMs., Comment: 44 pages; updated authors list and fixed author names

Conventional GAN-based models for talking head generation often suffer from limited quality and unstable training. Recent approaches based on diffusion models aimed to address these limitations and improve fidelity. However, they still face challenges, including extensive sampling times and difficulties in maintaining temporal consistency due to the high stochasticity of diffusion models. To overcome these challenges, we propose a novel motion-disentangled diffusion model for high-quality talking head generation, dubbed MoDiTalker. We introduce the two modules: audio-to-motion (AToM), designed to generate a synchronized lip motion from audio, and motion-to-video (MToV), designed to produce high-quality head video following the generated motion. AToM excels in capturing subtle lip movements by leveraging an audio attention mechanism. In addition, MToV enhances temporal consistency by leveraging an efficient tri-plane representation. Our experiments conducted on standard benchmarks demonstrate that our model achieves superior performance compared to existing models. We also provide comprehensive ablation studies and user study results.

Generating high-quality videos that synthesize desired realistic content is a challenging task due to their intricate high-dimensionality and complexity of videos. Several recent diffusion-based methods have shown comparable performance by compressing videos to a lower-dimensional latent space, using traditional video autoencoder architecture. However, such method that employ standard frame-wise 2D and 3D convolution fail to fully exploit the spatio-temporal nature of videos. To address this issue, we propose a novel hybrid video diffusion model, called HVDM, which can capture spatio-temporal dependencies more effectively. The HVDM is trained by a hybrid video autoencoder which extracts a disentangled representation of the video including: (i) a global context information captured by a 2D projected latent (ii) a local volume information captured by 3D convolutions with wavelet decomposition (iii) a frequency information for improving the video reconstruction. Based on this disentangled representation, our hybrid autoencoder provide a more comprehensive video latent enriching the generated videos with fine structures and details. Experiments on video generation benchamarks (UCF101, SkyTimelapse, and TaiChi) demonstrate that the proposed approach achieves state-of-the-art video generation quality, showing a wide range of video applications (e.g., long video generation, image-to-video, and video dynamics control)., Comment: Project page is available at https://hxngiee.github.io/HVDM/

There are two de facto standard architectures in recent computer vision: Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs). Strong inductive biases of convolutions help the model learn sample effectively, but such strong biases also limit the upper bound of CNNs when sufficient data are available. On the contrary, ViT is inferior to CNNs for small data but superior for sufficient data. Recent approaches attempt to combine the strengths of these two architectures. However, we show these approaches overlook that the optimal inductive bias also changes according to the target data scale changes by comparing various models' accuracy on subsets of sampled ImageNet at different ratios. In addition, through Fourier analysis of feature maps, the model's response patterns according to signal frequency changes, we observe which inductive bias is advantageous for each data scale. The more convolution-like inductive bias is included in the model, the smaller the data scale is required where the ViT-like model outperforms the ResNet performance. To obtain a model with flexible inductive bias on the data scale, we show reparameterization can interpolate inductive bias between convolution and self-attention. By adjusting the number of epochs the model stays in the convolution, we show that reparameterization from convolution to self-attention interpolates the Fourier analysis pattern between CNNs and ViTs. Adapting these findings, we propose Progressive Reparameterization Scheduling (PRS), in which reparameterization adjusts the required amount of convolution-like or self-attention-like inductive bias per layer. For small-scale datasets, our PRS performs reparameterization from convolution to self-attention linearly faster at the late stage layer. PRS outperformed previous studies on the small-scale dataset, e.g., CIFAR-100., Comment: Accepted at VIPriors ECCVW 2022, camera-ready version

Existing techniques for image-to-image translation commonly have suffered from two critical problems: heavy reliance on per-sample domain annotation and/or inability of handling multiple attributes per image. Recent truly-unsupervised methods adopt clustering approaches to easily provide per-sample one-hot domain labels. However, they cannot account for the real-world setting: one sample may have multiple attributes. In addition, the semantics of the clusters are not easily coupled to the human understanding. To overcome these, we present a LANguage-driven Image-to-image Translation model, dubbed LANIT. We leverage easy-to-obtain candidate attributes given in texts for a dataset: the similarity between images and attributes indicates per-sample domain labels. This formulation naturally enables multi-hot label so that users can specify the target domain with a set of attributes in language. To account for the case that the initial prompts are inaccurate, we also present prompt learning. We further present domain regularization loss that enforces translated images be mapped to the corresponding domain. Experiments on several standard benchmarks demonstrate that LANIT achieves comparable or superior performance to existing models., Comment: Accepted to CVPR 2023. Project Page: https://ku-cvlab.github.io/LANIT/

We present a novel Transformer-based network architecture for instance-aware image-to-image translation, dubbed InstaFormer, to effectively integrate global- and instance-level information. By considering extracted content features from an image as tokens, our networks discover global consensus of content features by considering context information through a self-attention module in Transformers. By augmenting such tokens with an instance-level feature extracted from the content feature with respect to bounding box information, our framework is capable of learning an interaction between object instances and the global image, thus boosting the instance-awareness. We replace layer normalization (LayerNorm) in standard Transformers with adaptive instance normalization (AdaIN) to enable a multi-modal translation with style codes. In addition, to improve the instance-awareness and translation quality at object regions, we present an instance-level content contrastive loss defined between input and translated image. We conduct experiments to demonstrate the effectiveness of our InstaFormer over the latest methods and provide extensive ablation studies., Comment: Accepted to CVPR 2022

Increasing attention has been drawn to the sentiment analysis of financial documents. The most popular examples of such documents include analyst reports and economic news, the analysis of which is frequently used to capture the trends in market sentiments. On the other hand, the significance of the role sentiment analysis plays in the financial domain has given rise to the efforts to construct a financial domain-specific sentiment lexicon. Sentiment lexicons lend a hand for solving various text mining tasks, such as unsupervised classification of text data, while alleviating the arduous human labor required for manual labeling. One of the challenges in the construction of an effective sentiment lexicon is that the semantic orientation of a word may change depending on the context in which it appears. For instance, the word ``profit" usually conveys positive sentiments; however, when the word is juxtaposed with another word ``decrease," the sentiment associated with the phrase ``profit decreases" now becomes negative. Hence, the sentiment of a given word may shift as one begins to consider the context surrounding the word. In this paper, we address this issue by incorporating context when building sentiment lexicon from a given corpus. Specifically, we construct a lexicon named Senti-DD for the Sentiment lexicon composed of Direction-Dependent words, which expresses each term a pair of a directional word and a direction-dependent word. Experiment results show that higher classification performance is achieved with Senti-DD, proving the effectiveness of our method for automatically constructing a context-aware sentiment lexicon in the financial domain.

The world today is experiencing an abundance of music like no other time, and attempts to group music into clusters have become increasingly prevalent. Common standards for grouping music were songs, artists, and genres, with artists or songs exploring similar genres of music seen as related. These clustering attempts serve critical purposes for various stakeholders involved in the music industry. For end users of music services, they may want to group their music so that they can easily navigate inside their music library; for music streaming platforms like Spotify, companies may want to establish a solid dataset of related songs in order to successfully provide personalized music recommendations and coherent playlists to their users. Due to increased competition in the streaming market, platforms are trying their best to find novel ways of learning similarities between audio to gain competitive advantage. Our team, comprised of music lovers with different tastes, was interested in the same issue, and created Music-Circles, an interactive visualization of music from the Billboard. Music-Circles links audio feature data offered by Spotify to popular songs to create unique vectors for each song, and calculate similarities between these vectors to cluster them. Through interacting with Music-Circles, users can gain understandings of audio features, view characteristic trends in popular music, and find out which music cluster they belong to.