Our paper aims to generate diverse and realistic ani-mal motion sequences from textual descriptions, without a large-scale animal text-motion dataset. While the task of text-driven human motion synthesis is already extensively studied and benchmarked, it remains challenging to transfer this success to other skeleton structures with limited data. In this work, we design a model architecture that imitates Generative Pretraining Transformer (GPT), utilizing prior knowledge learned from human data to the animal domain. We jointly train motion autoencoders for both animal and human motions and at the same time optimize through the similarity scores among human motion encoding, animal motion encoding, and text CLIP embedding. Presenting the first solution to this problem, we are able to generate animal motions with high diversity and fidelity, quantitatively and qualitatively outperforming the results of training human motion generation baselines on animal data. Additionally, we introduce AnimalML3D, the first text-animal motion dataset with 1240 animation sequences spanning 36 different ani-mal identities. We hope this dataset would mediate the data scarcity problem in text-driven animal motion generation, providing a new playground for the research community.

Recent advances in 3D avatar generation have gained significant attention. These breakthroughs aim to produce more realistic animatable avatars, narrowing the gap between virtual and real-world experiences. Most of existing works employ Score Distillation Sampling (SDS) loss, combined with a differentiable renderer and text condition, to guide a diffusion model in generating 3D avatars. How-ever, SDS often generates over-smoothed results with few facial details, thereby lacking the diversity compared with ancestral sampling. On the other hand, other works gen-erate 3D avatar from a single image, where the challenges of unwanted lighting effects, perspective views, and inferior image quality make them difficult to reliably reconstruct the 3D face meshes with the aligned complete textures. In this paper, we propose a novel 3D avatar generation approach termed UltrAvatar with enhanced fidelity of geometry, and superior quality of physically based rendering (PBR)textures without unwanted lighting. To this end, the proposed approach presents a diffuse color extraction model and an authenticity guided texture diffusion model. The former removes the unwanted lighting effects to reveal true diffuse colors, so that the generated avatars can be rendered under various lighting conditions. The latter follows two gradient-based guidances for generating PBR textures to render diverse face-identity features and details better aligning with 3D mesh geometry. We demonstrate the effectiveness and robustness of the proposed method, outperforming the state-of-the-art methods by a large margin in the experiments.

We consider the design of low-delay joint source-channel coding (JSCC) schemes for the transmission of discrete-time analog sources over noisy channels based on deep neural networks. The design problem is addressed as optimization of an autoencoder model, and several scenarios are discussed. For point-to-point communication of independent and identically distributed (i.i.d) Gaussian sources and Gauss-Markov sources over additive-white Gaussian noise (AWGN) channels, the encoder and decoder are constructed using recurrent neural networks (RNNs). With minimum prior knowledge used for design, the performance of these RNNs-based models is optimized using fine tuning techniques during training. Sinusoidal representation networks (SIRENs)-based models are proposed and optimized for three JSCC problems namely, transmitting multivariate Gaussian sources over AWGN channels, transmitting i.i.d Gaussian sources with side information at the decoder, and for communicating correlated sources over orthogonal Gaussian channels. We show that these deep learning-based JSCC schemes perform comparably or better than state-of-the-art (SOTA) traditional schemes. The proposed scheme can extend flexibly to different pairs of source and channel dimensions. Moreover, the spontaneously learned encoder mappings exhibit structured patterns that are interpretable.

Semi-supervised dense prediction tasks, such as semantic segmentation, can be greatly improved through the use of contrastive learning. However, this approach presents two key challenges: selecting informative negative samples from a highly redundant pool and implementing effective data augmentation. To address these challenges, we present an adversarial contrastive learning method specifically for semi-supervised semantic segmentation. Direct learning of adversarial negatives is adopted to retain discriminative information from the past, leading to higher learning efficiency. Our approach also leverages an advanced data augmentation strategy called AdverseMix, which combines information from under-performing classes to generate more diverse and challenging samples. Additionally, we use auxiliary labels and classifiers to prevent over-adversarial negatives from affecting the learning process. Our experiments on the Pascal VOC and Cityscapes datasets demonstrate that our method outperforms the state-of-the-art by a significant margin, even when using a small fraction of labeled data.

We revisit the joint source-channel coding (JSCC) problem of transmitting correlated sources over the additive white Gaussian noise channel using deep learning methods. Specifically, we consider the design of JSCC schemes for transmitting multivariate Gaussian sources, and Gauss-Markov processes over noisy channels with bandwidth (BW) compression and low delay. We show that encoding and decoding schemes represented by deep neural networks can be optimized jointly to obtain good JSCC schemes. Specifically, we adopt sinusoidal representation networks (SIRENs) for the transmission of multivariate Gaussian sources. The new architecture not only provides similar performance as the state-of-the-art (SOTA) with higher flexibility, but also results in interpretable encoder mappings. For the transmission of Gauss-Markov sources, recurrent neural networks (RNNs) are implemented to extract temporal information without resorting to explicit decorrelation prior to transmission. Experimental results show improved performance compared with traditional schemes.