Recent works demonstrate that using reinforcement learning (RL) with quality rewards can enhance the quality of generated images in text-to-image (T2I) generation. However, a simple aggregation of multiple rewards may cause over-optimization in certain metrics and degradation in others, and it is challenging to manually find the optimal weights. An effective strategy to jointly optimize multiple rewards in RL for T2I generation is highly desirable. This paper introduces Parrot, a novel multi-reward RL framework for T2I generation. Through the use of the batch-wise Pareto optimal selection, Parrot automatically identifies the optimal trade-off among different rewards during the RL optimization of the T2I generation. Additionally, Parrot employs a joint optimization approach for the T2I model and the prompt expansion network, facilitating the generation of quality-aware text prompts, thus further enhancing the final image quality. To counteract the potential catastrophic forgetting of the original user prompt due to prompt expansion, we introduce original prompt centered guidance at inference time, ensuring that the generated image remains faithful to the user input. Extensive experiments and a user study demonstrate that Parrot outperforms several baseline methods across various quality criteria, including aesthetics, human preference, image sentiment, and text-image alignment.
This report introduces a new family of multimodal models, Gemini, that exhibit remarkable capabilities across image, audio, video, and text understanding. The Gemini family consists of Ultra, Pro, and Nano sizes, suitable for applications ranging from complex reasoning tasks to on-device memory-constrained use-cases. Evaluation on a broad range of benchmarks shows that our most-capable Gemini Ultra model advances the state of the art in 30 of 32 of these benchmarks - notably being the first model to achieve human-expert performance on the well-studied exam benchmark MMLU, and improving the state of the art in every one of the 20 multimodal benchmarks we examined. We believe that the new capabilities of Gemini models in cross-modal reasoning and language understanding will enable a wide variety of use cases and we discuss our approach toward deploying them responsibly to users.
Recent Text-to-Image (T2I) generation models such as Stable Diffusion and Imagen have made significant progress in generating high-resolution images based on text descriptions. However, many generated images still suffer from issues such as artifacts/implausibility, misalignment with text descriptions, and low aesthetic quality. Inspired by the success of Reinforcement Learning with Human Feedback (RLHF) for large language models, prior works collected human-provided scores as feedback on generated images and trained a reward model to improve the T2I generation. In this paper, we enrich the feedback signal by (i) marking image regions that are implausible or misaligned with the text, and (ii) annotating which words in the text prompt are misrepresented or missing on the image. We collect such rich human feedback on 18K generated images and train a multimodal transformer to predict the rich feedback automatically. We show that the predicted rich human feedback can be leveraged to improve image generation, for example, by selecting high-quality training data to finetune and improve the generative models, or by creating masks with predicted heatmaps to inpaint the problematic regions. Notably, the improvements generalize to models (Muse) beyond those used to generate the images on which human feedback data were collected (Stable Diffusion variants).
Model initialization techniques are vital for improving the performance and reliability of deep learning models in medical computer vision applications. While much literature exists on non-medical images, the impacts on medical images, particularly chest X-rays (CXRs) are less understood. Addressing this gap, our study explores three deep model initialization techniques: Cold-start, Warm-start, and Shrink and Perturb start, focusing on adult and pediatric populations. We specifically focus on scenarios with periodically arriving data for training, thereby embracing the real-world scenarios of ongoing data influx and the need for model updates. We evaluate these models for generalizability against external adult and pediatric CXR datasets. We also propose novel ensemble methods: F-score-weighted Sequential Least-Squares Quadratic Programming (F-SLSQP) and Attention-Guided Ensembles with Learnable Fuzzy Softmax to aggregate weight parameters from multiple models to capitalize on their collective knowledge and complementary representations. We perform statistical significance tests with 95% confidence intervals and p-values to analyze model performance. Our evaluations indicate models initialized with ImageNet-pre-trained weights demonstrate superior generalizability over randomly initialized counterparts, contradicting some findings for non-medical images. Notably, ImageNet-pretrained models exhibit consistent performance during internal and external testing across different training scenarios. Weight-level ensembles of these models show significantly higher recall (p<0.05) during testing compared to individual models. Thus, our study accentuates the benefits of ImageNet-pretrained weight initialization, especially when used with weight-level ensembles, for creating robust and generalizable deep learning solutions.
Deep learning (DL) has demonstrated its innate capacity to independently learn hierarchical features from complex and multi-dimensional data. A common understanding is that its performance scales up with the amount of training data. Another data attribute is the inherent variety. It follows, therefore, that semantic redundancy, which is the presence of similar or repetitive information, would tend to lower performance and limit generalizability to unseen data. In medical imaging data, semantic redundancy can occur due to the presence of multiple images that have highly similar presentations for the disease of interest. Further, the common use of augmentation methods to generate variety in DL training may be limiting performance when applied to semantically redundant data. We propose an entropy-based sample scoring approach to identify and remove semantically redundant training data. We demonstrate using the publicly available NIH chest X-ray dataset that the model trained on the resulting informative subset of training data significantly outperforms the model trained on the full training set, during both internal (recall: 0.7164 vs 0.6597, p<0.05) and external testing (recall: 0.3185 vs 0.2589, p<0.05). Our findings emphasize the importance of information-oriented training sample selection as opposed to the conventional practice of using all available training data.
Human-Object Interaction (HOI) detection is a challenging computer vision task that requires visual models to address the complex interactive relationship between humans and objects and predict HOI triplets. Despite the challenges posed by the numerous interaction combinations, they also offer opportunities for multimodal learning of visual texts. In this paper, we present a systematic and unified framework (RmLR) that enhances HOI detection by incorporating structured text knowledge. Firstly, we qualitatively and quantitatively analyze the loss of interaction information in the two-stage HOI detector and propose a re-mining strategy to generate more comprehensive visual representation.Secondly, we design more fine-grained sentence- and word-level alignment and knowledge transfer strategies to effectively address the many-to-many matching problem between multiple interactions and multiple texts.These strategies alleviate the matching confusion problem that arises when multiple interactions occur simultaneously, thereby improving the effectiveness of the alignment process. Finally, HOI reasoning by visual features augmented with textual knowledge substantially improves the understanding of interactions. Experimental results illustrate the effectiveness of our approach, where state-of-the-art performance is achieved on public benchmarks. We further analyze the effects of different components of our approach to provide insights into its efficacy.
Most existing studies on massive grant-free access, proposed to support massive machine-type communications (mMTC) for the Internet of things (IoT), assume Rayleigh fading and perfect synchronization for simplicity. However, in practice, line-of-sight (LoS) components generally exist, and time and frequency synchronization are usually imperfect. This paper systematically investigates maximum likelihood estimation (MLE)-based device activity detection under Rician fading for massive grant-free access with perfect and imperfect synchronization. Specifically, we formulate device activity detection in the synchronous case and joint device activity and offset detection in three asynchronous cases (i.e., time, frequency, and time and frequency asynchronous cases) as MLE problems. In the synchronous case, we propose an iterative algorithm to obtain a stationary point of the MLE problem. In each asynchronous case, we propose two iterative algorithms with identical detection performance but different computational complexities. In particular, one is computationally efficient for small ranges of offsets, whereas the other one, relying on fast Fourier transform (FFT) and inverse FFT, is computationally efficient for large ranges of offsets. The proposed algorithms generalize the existing MLE-based methods for Rayleigh fading and perfect synchronization. Numerical results show the notable gains of the proposed algorithms over existing methods in detection accuracy and computation time.
Neural networks (NN) have demonstrated remarkable capabilities in various tasks, but their computation-intensive nature demands faster and more energy-efficient hardware implementations. Optics-based platforms, using technologies such as silicon photonics and spatial light modulators, offer promising avenues for achieving this goal. However, training multiple trainable layers in tandem with these physical systems poses challenges, as they are difficult to fully characterize and describe with differentiable functions, hindering the use of error backpropagation algorithm. The recently introduced Forward-Forward Algorithm (FFA) eliminates the need for perfect characterization of the learning system and shows promise for efficient training with large numbers of programmable parameters. The FFA does not require backpropagating an error signal to update the weights, rather the weights are updated by only sending information in one direction. The local loss function for each set of trainable weights enables low-power analog hardware implementations without resorting to metaheuristic algorithms or reinforcement learning. In this paper, we present an experiment utilizing multimode nonlinear wave propagation in an optical fiber demonstrating the feasibility of the FFA approach using an optical system. The results show that incorporating optical transforms in multilayer NN architectures trained with the FFA, can lead to performance improvements, even with a relatively small number of trainable weights. The proposed method offers a new path to the challenge of training optical NNs and provides insights into leveraging physical transformations for enhancing NN performance.
Automatic and accurate segmentation of aortic vessel tree (AVT) in computed tomography (CT) scans is crucial for early detection, diagnosis and prognosis of aortic diseases, such as aneurysms, dissections and stenosis. However, this task remains challenges, due to the complexity of aortic vessel tree and amount of CT angiography data. In this technical report, we use two-stage fully convolutional networks (FCNs) to automatically segment AVT in CTA scans from multiple centers. Specifically, we firstly adopt a 3D FCN with U-shape network architecture to segment AVT in order to produce topology attention and accelerate medical image analysis pipeline. And then another one 3D FCN is trained to segment branches of AVT along the pseudo-centerline of AVT. In the 2023 MICCAI Segmentation of the Aorta (SEG.A.) Challenge , the reported method was evaluated on the public dataset of 56 cases. The resulting Dice Similarity Coefficient (DSC) is 0.920, Jaccard Similarity Coefficient (JSC) is 0.861, Recall is 0.922, and Precision is 0.926 on a 5-fold random split of training and validation set.
We propose a method for adding sound-guided visual effects to specific regions of videos with a zero-shot setting. Animating the appearance of the visual effect is challenging because each frame of the edited video should have visual changes while maintaining temporal consistency. Moreover, existing video editing solutions focus on temporal consistency across frames, ignoring the visual style variations over time, e.g., thunderstorm, wave, fire crackling. To overcome this limitation, we utilize temporal sound features for the dynamic style. Specifically, we guide denoising diffusion probabilistic models with an audio latent representation in the audio-visual latent space. To the best of our knowledge, our work is the first to explore sound-guided natural video editing from various sound sources with sound-specialized properties, such as intensity, timbre, and volume. Additionally, we design optical flow-based guidance to generate temporally consistent video frames, capturing the pixel-wise relationship between adjacent frames. Experimental results show that our method outperforms existing video editing techniques, producing more realistic visual effects that reflect the properties of sound. Please visit our page: https://kuai-lab.github.io/soundini-gallery/.