Large language models (LLMs) have gained popularity recently due to their outstanding performance in various downstream Natural Language Processing (NLP) tasks. However, low-resource languages are still lagging behind current state-of-the-art (SOTA) developments in the field of NLP due to insufficient resources to train LLMs. Ethiopian languages exhibit remarkable linguistic diversity, encompassing a wide array of scripts, and are imbued with profound religious and cultural significance. This paper introduces EthioLLM -- multilingual large language models for five Ethiopian languages (Amharic, Ge'ez, Afan Oromo, Somali, and Tigrinya) and English, and Ethiobenchmark -- a new benchmark dataset for various downstream NLP tasks. We evaluate the performance of these models across five downstream NLP tasks. We open-source our multilingual language models, new benchmark datasets for various downstream tasks, and task-specific fine-tuned language models and discuss the performance of the models. Our dataset and models are available at the https://huggingface.co/EthioNLP repository.
ORCA (Shen et al., 2023) is a recent technique for cross-modal fine-tuning, i.e., applying pre-trained transformer models to modalities beyond their training data. The technique consists primarily of training an embedder and fine-tuning the embedder and model. Despite its high performance on a variety of downstream tasks, we do not understand precisely how each of these components contribute to ORCA's success. Therefore, we run a series of ablations and find that embedder training does not help 2D tasks at all, contrary to what the original paper posits. In 1D tasks, some amount of embedder training is necessary but more is not better. In 4 out of 6 datasets we experiment with, it is model fine-tuning that makes the biggest difference. Through our ablations and baselines, we contribute a better understanding of the individual components of ORCA.
Current approaches for 3D human motion synthesis can generate high-quality 3D animations of digital humans performing a wide variety of actions and gestures. However, there is still a notable technological gap in addressing the complex dynamics of multi-human interactions within this paradigm. In this work, we introduce ReMoS, a denoising diffusion-based probabilistic model for reactive motion synthesis that explores two-person interactions. Given the motion of one person, we synthesize the reactive motion of the second person to complete the interactions between the two. In addition to synthesizing the full-body motions, we also synthesize plausible hand interactions. We show the performance of ReMoS under a wide range of challenging two-person scenarios including pair-dancing, Ninjutsu, kickboxing, and acrobatics, where one person's movements have complex and diverse influences on the motions of the other. We further propose the ReMoCap dataset for two-person interactions consisting of full-body and hand motions. We evaluate our approach through multiple quantitative metrics, qualitative visualizations, and a user study. Our results are usable in interactive applications while also providing an adequate amount of control for animators.
Virtual character animation and movement synthesis have advanced rapidly during recent years, especially through a combination of extensive motion capture datasets and machine learning. A remaining challenge is interactively simulating characters that fatigue when performing extended motions, which is indispensable for the realism of generated animations. However, capturing such movements is problematic, as performing movements like backflips with fatigued variations up to exhaustion raises capture cost and risk of injury. Surprisingly, little research has been done on faithful fatigue modeling. To address this, we propose a deep reinforcement learning-based approach, which -- for the first time in literature -- generates control policies for full-body physically simulated agents aware of cumulative fatigue. For this, we first leverage Generative Adversarial Imitation Learning (GAIL) to learn an expert policy for the skill; Second, we learn a fatigue policy by limiting the generated constant torque bounds based on endurance time to non-linear, state- and time-dependent limits in the joint-actuation space using a Three-Compartment Controller (3CC) model. Our results demonstrate that agents can adapt to different fatigue and rest rates interactively, and discover realistic recovery strategies without the need for any captured data of fatigued movement.
Video super-resolution (VSR) techniques, especially deep-learning-based algorithms, have drastically improved over the last few years and shown impressive performance on synthetic data. However, their performance on real-world video data suffers because of the complexity of real-world degradations and misaligned video frames. Since obtaining a synthetic dataset consisting of low-resolution (LR) and high-resolution (HR) frames are easier than obtaining real-world LR and HR images, in this paper, we propose synthesizing real-world degradations on synthetic training datasets. The proposed synthetic real-world degradations (SRWD) include a combination of the blur, noise, downsampling, pixel binning, and image and video compression artifacts. We then propose using a random shuffling-based strategy to simulate these degradations on the training datasets and train a single end-to-end deep neural network (DNN) on the proposed larger variation of realistic synthesized training data. Our quantitative and qualitative comparative analysis shows that the proposed training strategy using diverse realistic degradations improves the performance by 7.1 % in terms of NRQM compared to RealBasicVSR and by 3.34 % compared to BSRGAN on the VideoLQ dataset. We also introduce a new dataset that contains high-resolution real-world videos that can serve as a common ground for bench-marking.
Video depth estimation is crucial in various applications, such as scene reconstruction and augmented reality. In contrast to the naive method of estimating depths from images, a more sophisticated approach uses temporal information, thereby eliminating flickering and geometrical inconsistencies. We propose a consistent method for dense video depth estimation; however, unlike the existing monocular methods, ours relates to stereo videos. This technique overcomes the limitations arising from the monocular input. As a benefit of using stereo inputs, a left-right consistency loss is introduced to improve the performance. Besides, we use SLAM-based camera pose estimation in the process. To address the problem of depth blurriness during test-time training (TTT), we present an edge-preserving loss function that improves the visibility of fine details while preserving geometrical consistency. We show that our edge-aware stereo video model can accurately estimate the dense depth maps.
Accurate depth maps are essential in various applications, such as autonomous driving, scene reconstruction, point-cloud creation, etc. However, monocular-depth estimation (MDE) algorithms often fail to provide enough texture & sharpness, and also are inconsistent for homogeneous scenes. These algorithms mostly use CNN or vision transformer-based architectures requiring large datasets for supervised training. But, MDE algorithms trained on available depth datasets do not generalize well and hence fail to perform accurately in diverse real-world scenes. Moreover, the ground-truth depth maps are either lower resolution or sparse leading to relatively inconsistent depth maps. In general, acquiring a high-resolution ground truth dataset with pixel-level precision for accurate depth prediction is an expensive, and time-consuming challenge. In this paper, we generate a high-resolution synthetic depth dataset (HRSD) of dimension 1920 X 1080 from Grand Theft Auto (GTA-V), which contains 100,000 color images and corresponding dense ground truth depth maps. The generated datasets are diverse and have scenes from indoors to outdoors, from homogeneous surfaces to textures. For experiments and analysis, we train the DPT algorithm, a state-of-the-art transformer-based MDE algorithm on the proposed synthetic dataset, which significantly increases the accuracy of depth maps on different scenes by 9 %. Since the synthetic datasets are of higher resolution, we propose adding a feature extraction module in the transformer encoder and incorporating an attention-based loss, further improving the accuracy by 15 %.
Memory efficiency is crucial in training deep learning networks on resource-restricted devices. During backpropagation, forward tensors are used to calculate gradients. Despite the option of keeping those dependencies in memory until they are reused in backpropagation, some forward tensors can be discarded and recomputed later from saved tensors, so-called checkpoints. This allows, in particular, for resource-constrained heterogeneous environments to make use of all available compute devices. Unfortunately, the definition of these checkpoints is a non-trivial problem and poses a challenge to the programmer - improper or excessive recomputations negate the benefit of checkpointing. In this article, we present XEngine, an approach that schedules network operators to heterogeneous devices in low memory environments by determining checkpoints and recomputations of tensors. Our approach selects suitable resources per timestep and operator and optimizes the end-to-end time for neural networks taking the memory limitation of each device into account. For this, we formulate a mixed-integer quadratic program (MIQP) to schedule operators of deep learning networks on heterogeneous systems. We compare our MIQP solver XEngine against Checkmate, a mixed-integer linear programming (MILP) approach that solves recomputation on a single device. Our solver finds solutions that are up to 22.5 % faster than the fastest Checkmate schedule in which the network is computed exclusively on a single device. We also find valid schedules for networks making use of both central processing units and graphics processing units if memory limitations do not allow scheduling exclusively to the graphics processing unit.
Can we make virtual characters in a scene interact with their surrounding objects through simple instructions? Is it possible to synthesize such motion plausibly with a diverse set of objects and instructions? Inspired by these questions, we present the first framework to synthesize the full-body motion of virtual human characters performing specified actions with 3D objects placed within their reach. Our system takes as input textual instructions specifying the objects and the associated intentions of the virtual characters and outputs diverse sequences of full-body motions. This is in contrast to existing work, where full-body action synthesis methods generally do not consider object interactions, and human-object interaction methods focus mainly on synthesizing hand or finger movements for grasping objects. We accomplish our objective by designing an intent-driven full-body motion generator, which uses a pair of decoupled conditional variational autoencoders (CVAE) to learn the motion of the body parts in an autoregressive manner. We also optimize for the positions of the objects with six degrees of freedom (6DoF) such that they plausibly fit within the hands of the synthesized characters. We compare our proposed method with the existing methods of motion synthesis and establish a new and stronger state-of-the-art for the task of intent-driven motion synthesis. Through a user study, we further show that our synthesized full-body motions appear more realistic to the participants in more than 80% of scenarios compared to the current state-of-the-art methods, and are perceived to be as good as the ground truth on several occasions.