Chao Tang

Ning Yang, Chao Tang, Yuhai Tu

Generalization is one of the most important problems in deep learning (DL). In the overparameterized regime in neural networks, there exist many low-loss solutions that fit the training data equally well. The key question is which solution is more generalizable. Empirical studies showed a strong correlation between flatness of the loss landscape at a solution and its generalizability, and stochastic gradient descent (SGD) is crucial in finding the flat solutions. To understand how SGD drives the learning system to flat solutions, we construct a simple model whose loss landscape has a continuous set of degenerate (or near degenerate) minima. By solving the Fokker-Planck equation of the underlying stochastic learning dynamics, we show that due to its strong anisotropy the SGD noise introduces an additional effective loss term that decreases with flatness and has an overall strength that increases with the learning rate and batch-to-batch variation. We find that the additional landscape-dependent SGD-loss breaks the degeneracy and serves as an effective regularization for finding flat solutions. Furthermore, a stronger SGD noise shortens the convergence time to the flat solutions. However, we identify an upper bound for the SGD noise beyond which the system fails to converge. Our results not only elucidate the role of SGD for generalization they may also have important implications for hyperparameter selection for learning efficiently without divergence.

Sixiang Chen, Jinbin Bai, Zhuoran Zhao et al.

The landscape of image generation has rapidly evolved, from early GAN-based approaches to diffusion models and, most recently, to unified generative architectures that seek to bridge understanding and generation tasks. Recent advances, especially the GPT-4o, have demonstrated the feasibility of high-fidelity multimodal generation, their architectural design remains mysterious and unpublished. This prompts the question of whether image and text generation have already been successfully integrated into a unified framework for those methods. In this work, we conduct an empirical study of GPT-4o's image generation capabilities, benchmarking it against leading open-source and commercial models. Our evaluation covers four main categories, including text-to-image, image-to-image, image-to-3D, and image-to-X generation, with more than 20 tasks. Our analysis highlights the strengths and limitations of GPT-4o under various settings, and situates it within the broader evolution of generative modeling. Through this investigation, we identify promising directions for future unified generative models, emphasizing the role of architectural design and data scaling. For a high-definition version of the PDF, please refer to the link on GitHub: \href{https://github.com/Ephemeral182/Empirical-Study-of-GPT-4o-Image-Gen}{https://github.com/Ephemeral182/Empirical-Study-of-GPT-4o-Image-Gen}.

Chao Tang, Anxing Xiao, Yuhong Deng et al.

Imitating tool manipulation from human videos offers an intuitive approach to teaching robots, while also providing a promising and scalable alternative to labor-intensive teleoperation data collection for visuomotor policy learning. While humans can mimic tool manipulation behavior by observing others perform a task just once and effortlessly transfer the skill to diverse tools for functionally equivalent tasks, current robots struggle to achieve this level of generalization. A key challenge lies in establishing function-level correspondences, considering the significant geometric variations among functionally similar tools, referred to as intra-function variations. To address this challenge, we propose MimicFunc, a framework that establishes functional correspondences with function frame, a function-centric local coordinate frame constructed with keypoint-based abstraction, for imitating tool manipulation skills. Experiments demonstrate that MimicFunc effectively enables the robot to generalize the skill from a single RGB-D human video to manipulating novel tools for functionally equivalent tasks. Furthermore, leveraging MimicFunc's one-shot generalization capability, the generated rollouts can be used to train visuomotor policies without requiring labor-intensive teleoperation data collection for novel objects. Our code and video are available at https://sites.google.com/view/mimicfunc.

Jianzong Wu, Chao Tang, Jingbo Wang et al.

Story visualization, the task of creating visual narratives from textual descriptions, has seen progress with text-to-image generation models. However, these models often lack effective control over character appearances and interactions, particularly in multi-character scenes. To address these limitations, we propose a new task: \textbf{customized manga generation} and introduce \textbf{DiffSensei}, an innovative framework specifically designed for generating manga with dynamic multi-character control. DiffSensei integrates a diffusion-based image generator with a multimodal large language model (MLLM) that acts as a text-compatible identity adapter. Our approach employs masked cross-attention to seamlessly incorporate character features, enabling precise layout control without direct pixel transfer. Additionally, the MLLM-based adapter adjusts character features to align with panel-specific text cues, allowing flexible adjustments in character expressions, poses, and actions. We also introduce \textbf{MangaZero}, a large-scale dataset tailored to this task, containing 43,264 manga pages and 427,147 annotated panels, supporting the visualization of varied character interactions and movements across sequential frames. Extensive experiments demonstrate that DiffSensei outperforms existing models, marking a significant advancement in manga generation by enabling text-adaptable character customization. The project page is https://jianzongwu.github.io/projects/diffsensei/.

Chao Tang, Arwa Dabbech, Adrian Jackson et al.

The next-generation radio-interferometric (RI) telescopes require imaging algorithms capable of forming high-resolution high-dynamic-range images from large data volumes spanning wide frequency bands. Recently, AIRI, a plug-and-play (PnP) approach taking the forward-backward algorithmic structure (FB), has demonstrated state-of-the-art performance in monochromatic RI imaging by alternating a data-fidelity step with a regularisation step via learned denoisers. In this work, we introduce HyperAIRI, its hyperspectral extension, underpinned by learned hyperspectral denoisers enforcing a power-law spectral model. For each spectral channel, the HyperAIRI denoiser takes as input its current image estimate, alongside estimates of its two immediate neighbouring channels and the spectral index map, and provides as output its associated denoised image. To ensure convergence of HyperAIRI, the denoisers are trained with a Jacobian regularisation enforcing non-expansiveness. To accommodate varying dynamic ranges, we assemble a shelf of pre-trained denoisers, each tailored to a specific dynamic range. At each HyperAIRI iteration, the spectral channels of the target image cube are updated in parallel using dynamic-range-matched denoisers from the pre-trained shelf. The denoisers are also endowed with a spatial image faceting functionality, enabling scalability to varied image sizes. Additionally, we formally introduce Hyper-uSARA, a variant of the optimisation-based algorithm HyperSARA, promoting joint sparsity across spectral channels via the l2,1-norm, also adopting FB. We evaluate HyperAIRI's performance on simulated and real observations. We showcase its superior performance compared to its optimisation-based counterpart Hyper-uSARA, CLEAN's hyperspectral variant in WSClean, and the monochromatic imaging algorithms AIRI and uSARA.

Matthieu Terris, Arwa Dabbech, Chao Tang et al.

We introduce a new class of iterative image reconstruction algorithms for radio interferometry, at the interface of convex optimization and deep learning, inspired by plug-and-play methods. The approach consists in learning a prior image model by training a deep neural network (DNN) as a denoiser, and substituting it for the handcrafted proximal regularization operator of an optimization algorithm. The proposed AIRI (``AI for Regularization in radio-interferometric Imaging'') framework, for imaging complex intensity structure with diffuse and faint emission from visibility data, inherits the robustness and interpretability of optimization, and the learning power and speed of networks. Our approach relies on three steps. Firstly, we design a low dynamic range training database from optical intensity images. Secondly, we train a DNN denoiser at a noise level inferred from the signal-to-noise ratio of the data. We use training losses enhanced with a nonexpansiveness term ensuring algorithm convergence, and including on-the-fly database dynamic range enhancement via exponentiation. Thirdly, we plug the learned denoiser into the forward-backward optimization algorithm, resulting in a simple iterative structure alternating a denoising step with a gradient-descent data-fidelity step. We have validated AIRI against CLEAN, optimization algorithms of the SARA family, and a DNN trained to reconstruct the image directly from visibility data. Simulation results show that AIRI is competitive in imaging quality with SARA and its unconstrained forward-backward-based version uSARA, while providing significant acceleration. CLEAN remains faster but offers lower quality. The end-to-end DNN offers further acceleration, but with far lower quality than AIRI.

Yunzhi Lin, Chao Tang, Fu-Jen Chu et al.

Shape informs how an object should be grasped, both in terms of where and how. As such, this paper describes a segmentation-based architecture for decomposing objects sensed with a depth camera into multiple primitive shapes, along with a post-processing pipeline for robotic grasping. Segmentation employs a deep network, called PS-CNN, trained on synthetic data with 6 classes of primitive shapes and generated using a simulation engine. Each primitive shape is designed with parametrized grasp families, permitting the pipeline to identify multiple grasp candidates per shape region. The grasps are rank ordered, with the first feasible one chosen for execution. For task-free grasping of individual objects, the method achieves a 94.2% success rate placing it amongst the top performing grasp methods when compared to top-down and SE(3)-based approaches. Additional tests involving variable viewpoints and clutter demonstrate robustness to setup. For task-oriented grasping, PS-CNN achieves a 93.0% success rate. Overall, the outcomes support the hypothesis that explicitly encoding shape primitives within a grasping pipeline should boost grasping performance, including task-free and task-relevant grasp prediction.