Sheng Quan Xie

Jaskirat Singh, Xingjian Leng, Zongze Wu et al.

Representation alignment (REPA) guides generative training by distilling representations from a strong, pretrained vision encoder to intermediate diffusion features. We investigate a fundamental question: what aspect of the target representation matters for generation, its \textit{global} \revision{semantic} information (e.g., measured by ImageNet-1K accuracy) or its spatial structure (i.e. pairwise cosine similarity between patch tokens)? Prevalent wisdom holds that stronger global semantic performance leads to better generation as a target representation. To study this, we first perform a large-scale empirical analysis across 27 different vision encoders and different model scales. The results are surprising; spatial structure, rather than global performance, drives the generation performance of a target representation. To further study this, we introduce two straightforward modifications, which specifically accentuate the transfer of \emph{spatial} information. We replace the standard MLP projection layer in REPA with a simple convolution layer and introduce a spatial normalization layer for the external representation. Surprisingly, our simple method (implemented in $<$4 lines of code), termed iREPA, consistently improves convergence speed of REPA, across a diverse set of vision encoders, model sizes, and training variants (such as REPA, REPA-E, Meanflow, JiT etc). %, etc. Our work motivates revisiting the fundamental working mechanism of representational alignment and how it can be leveraged for improved training of generative models. The code and project page are available at https://end2end-diffusion.github.io/irepa

Xichen Pan, Satya Narayan Shukla, Aashu Singh et al.

Unified multimodal models aim to integrate understanding (text output) and generation (pixel output), but aligning these different modalities within a single architecture often demands complex training recipes and careful data balancing. We introduce MetaQueries, a set of learnable queries that act as an efficient interface between autoregressive multimodal LLMs (MLLMs) and diffusion models. MetaQueries connects the MLLM's latents to the diffusion decoder, enabling knowledge-augmented image generation by leveraging the MLLM's deep understanding and reasoning capabilities. Our method simplifies training, requiring only paired image-caption data and standard diffusion objectives. Notably, this transfer is effective even when the MLLM backbone remains frozen, thereby preserving its state-of-the-art multimodal understanding capabilities while achieving strong generative performance. Additionally, our method is flexible and can be easily instruction-tuned for advanced applications such as image editing and subject-driven generation.

David Fan, Shengbang Tong, Jiachen Zhu et al.

Visual Self-Supervised Learning (SSL) currently underperforms Contrastive Language-Image Pretraining (CLIP) in multimodal settings such as Visual Question Answering (VQA). This multimodal gap is often attributed to the semantics introduced by language supervision, even though visual SSL and CLIP models are often trained on different data. In this work, we ask the question: "Do visual self-supervised approaches lag behind CLIP due to the lack of language supervision, or differences in the training data?" We study this question by training both visual SSL and CLIP models on the same MetaCLIP data, and leveraging VQA as a diverse testbed for vision encoders. In this controlled setup, visual SSL models scale better than CLIP models in terms of data and model capacity, and visual SSL performance does not saturate even after scaling up to 7B parameters. Consequently, we observe visual SSL methods achieve CLIP-level performance on a wide range of VQA and classic vision benchmarks. These findings demonstrate that pure visual SSL can match language-supervised visual pretraining at scale, opening new opportunities for vision-centric representation learning.

Senwei Xie, Hongyu Wang, Zhanqi Xiao et al.

Zero-shot generalization across various robots, tasks and environments remains a significant challenge in robotic manipulation. Policy code generation methods use executable code to connect high-level task descriptions and low-level action sequences, leveraging the generalization capabilities of large language models and atomic skill libraries. In this work, we propose Robotic Programmer (RoboPro), a robotic foundation model, enabling the capability of perceiving visual information and following free-form instructions to perform robotic manipulation with policy code in a zero-shot manner. To address low efficiency and high cost in collecting runtime code data for robotic tasks, we devise Video2Code to synthesize executable code from extensive videos in-the-wild with off-the-shelf vision-language model and code-domain large language model. Extensive experiments show that RoboPro achieves the state-of-the-art zero-shot performance on robotic manipulation in both simulators and real-world environments. Specifically, the zero-shot success rate of RoboPro on RLBench surpasses the state-of-the-art model GPT-4o by 11.6%, which is even comparable to a strong supervised training baseline. Furthermore, RoboPro is robust to variations on API formats and skill sets.

Huaihai Lyu, Chaofan Chen, Senwei Xie et al.

Existing Vision-Language-Action (VLA) models can be broadly categorized into diffusion-based and auto-regressive (AR) approaches: diffusion models capture continuous action distributions but rely on computationally heavy iterative denoising. In contrast, AR models enable efficient optimization and flexible sequence construction, making them better suited for large-scale pretraining. To further improve AR efficiency, particularly when action chunks induce extended and high-dimensional sequences, prior work applies entropy-guided and token-frequency techniques to shorten the sequence length. However, such compression struggled with \textit{poor reconstruction or inefficient compression}. Motivated by this, we introduce an Omni Swift Action Tokenizer, which learns a compact, transferable action representation. Specifically, we first normalize value ranges and temporal horizons to obtain a consistent representation with B-Spline encoding. Then, we apply multi-stage residual quantization to the position, rotation, and gripper subspaces, producing compressed discrete tokens with coarse-to-fine granularity for each part. After pre-training on the large-scale dataset Droid, the resulting discrete tokenization shortens the training sequence by 6.8$\times$, and lowers the target entropy. To further explore the potential of OmniSAT, we develop a cross-embodiment learning strategy that builds on the unified action-pattern space and jointly leverages robot and human demonstrations. It enables scalable auxiliary supervision from heterogeneous egocentric videos. Across diverse real-robot and simulation experiments, OmniSAT encompasses higher compression while preserving reconstruction quality, enabling faster AR training convergence and model performance.

Basar Kutukcu, Sinan Xie, Sabur Baidya et al.

Nvidia Jetson boards are powerful systems for executing artificial intelligence workloads in edge and mobile environments due to their effective GPU hardware and widely supported software stack. In addition to these benefits, Nvidia Jetson boards provide large configurability by giving the user the choice to modify many hardware parameters. This large space of configurability creates the need of searching the optimal configurations based on the user's requirements. In this work, we propose JExplore, a multi-board software and hardware design space exploration tool. JExplore can be integrated with any search tool, hence creating a common benchmarking ground for the search algorithms. Moreover, it accelerates the exploration of user application and Nvidia Jetson configurations for researchers and engineers by encapsulating host-client communication, configuration management, and metric measurement.

Tianxiao Gao, Wu Wei, Zhongbin Cai et al.

Monocular depth estimation and semantic segmentation are two fundamental goals of scene understanding. Due to the advantages of task interaction, many works study the joint task learning algorithm. However, most existing methods fail to fully leverage the semantic labels, ignoring the provided context structures and only using them to supervise the prediction of segmentation split, which limit the performance of both tasks. In this paper, we propose a network injected with contextual information (CI-Net) to solve the problem. Specifically, we introduce self-attention block in the encoder to generate attention map. With supervision from the ideal attention map created by semantic label, the network is embedded with contextual information so that it could understand scene better and utilize correlated features to make accurate prediction. Besides, a feature sharing module is constructed to make the task-specific features deeply fused and a consistency loss is devised to make the features mutually guided. We evaluate the proposed CI-Net on the NYU-Depth-v2 and SUN-RGBD datasets. The experimental results validate that our proposed CI-Net could effectively improve the accuracy of semantic segmentation and depth estimation.