ICLR 2025 Papers

Video generation technologies are developing rapidly and have broad potential applications. Among these technologies, camera control is crucial for generating professional-quality videos that accurately meet user expectations. However, existing camera control methods still suffer from several limitations, including control precision and the neglect of the control for subject motion dynamics. In this work, we propose I2VControl-Camera, a novel camera control method that significantly enhances controllability while providing adjustability over the strength of subject motion. To improve control precision, we employ point trajectory in the camera coordinate system instead of only extrinsic matrix information as our control signal. To accurately control and adjust the strength of subject motion, we explicitly model the higher-order components of the video trajectory expansion, not merely the linear terms, and design an operator that effectively represents the motion strength. We use an adapter architecture that is independent of the base model structure. Experiments on static and dynamic scenes show that our framework outperformances previous methods both quantitatively and qualitatively. Project page: https://wanquanf.github.io/I2VControlCamera.

This paper introduces StreamV2V, a diffusion model that achieves real-time streaming video-to-video (V2V) translation with user prompts. Unlike prior V2V methods using batches to process limited frames, we opt to process frames in a streaming fashion, to support unlimited frames. At the heart of StreamV2V lies a backward-looking principle that relates the present to the past. This is realized by maintaining a feature bank, which archives information from past frames. For incoming frames, StreamV2V extends self-attention to include banked keys and values, and directly fuses similar past features into the output. The feature bank is continually updated by merging stored and new features, making it compact yet informative. StreamV2V stands out for its adaptability and efficiency, seamlessly integrating with image diffusion models without fine-tuning. It can run 20 FPS on one A100 GPU, being 15$\times$, 46$\times$, 108$\times$, and 158$\times$ faster than FlowVid, CoDeF, Rerender, and TokenFlow, respectively. Quantitative metrics and user studies confirm StreamV2V's exceptional ability to maintain temporal consistency.

A recent line of work in mechanistic interpretability has focused on reverse-engineering the computation performed by neural networks trained on the binary operation of finite groups. We investigate the internals of one-hidden-layer neural networks trained on this task, revealing previously unidentified structure and producing a more complete description of such models in a step towards unifying the explanations of previous works (Chughtai et al., 2023; Stander et al., 2024). Notably, these models approximate equivariance in each input argument. We verify that our explanation applies to a large fraction of networks trained on this task by translating it into a compact proof of model performance, a quantitative evaluation of the extent to which we faithfully and concisely explain model internals. In the main text, we focus on the symmetric group S5. For models trained on this group, our explanation yields a guarantee of model accuracy that runs 3x faster than brute force and gives a >=95% accuracy bound for 45% of the models we trained. We were unable to obtain nontrivial non-vacuous accuracy bounds using only explanations from previous works.

Text-driven video generation has advanced significantly due to developments in diffusion models. Beyond the training and sampling phases, recent studies have investigated noise priors of diffusion models, as improved noise priors yield better generation results. One recent approach employs the Fourier transform to manipulate noise, marking the initial exploration of frequency operations in this context. However, it often generates videos that lack motion dynamics and imaging details. In this work, we provide a comprehensive theoretical analysis of the variance decay issue present in existing methods, contributing to the loss of details and motion dynamics. Recognizing the critical impact of noise distribution on generation quality, we introduce FreqPrior, a novel noise initialization strategy that refines noise in the frequency domain. Our method features a novel filtering technique designed to address different frequency signals while maintaining the noise prior distribution that closely approximates a standard Gaussian distribution. Additionally, we propose a partial sampling process by perturbing the latent at an intermediate timestep while finding the noise prior, significantly reducing inference time without compromising quality. Extensive experiments on VBench demonstrate that our method achieves the highest scores in both quality and semantic assessments, resulting in the best overall total score. These results highlight the superiority of our proposed noise prior.

We propose ReKV, a novel training-free approach that enables efficient streaming video question-answering (StreamingVQA), by seamlessly integrating with existing Video Large Language Models (Video-LLMs). Traditional VideoQA systems struggle with long videos, as they must process entire videos before responding to queries, and repeat this process for each new question. In contrast, our approach analyzes long videos in a streaming manner, allowing for prompt responses as soon as user queries are received. Building on a common Video-LLM, we first incorporate a sliding-window attention mechanism, ensuring that input frames attend to a limited number of preceding frames, thereby reducing computational overhead. To prevent information loss, we store processed video key-value caches (KV-Caches) in RAM and disk, reloading them into GPU memory as needed. Additionally, we introduce a retrieval method that leverages an external retriever or the parameters within Video-LLMs to retrieve only query-relevant KV-Caches, ensuring both efficiency and accuracy in question answering. ReKV enables the separation of video analyzing and question-answering across different processes and GPUs, significantly enhancing the efficiency of StreamingVQA. Through comprehensive experimentation, we validate the efficacy and practicality of our approach, which significantly boosts efficiency and enhances applicability over existing VideoQA models.

Understanding, navigating, and exploring the 3D physical real world has long been a central challenge in the development of artificial intelligence. In this work, we take a step toward this goal by introducing *GenEx*, a system capable of planning complex embodied world exploration, guided by its generative imagination that forms expectations about the surrounding environments. *GenEx* generates high-quality, continuous 360-degree virtual environments, achieving robust loop consistency and active 3D mapping over extended trajectories. Leveraging generative imagination, GPT-assisted agents can undertake complex embodied tasks, including goal-agnostic exploration and goal-driven navigation. Agents utilize imagined observations to update their beliefs, simulate potential outcomes, and enhance their decision-making. Training on the synthetic urban dataset *GenEx-DB* and evaluation on *GenEx-EQA* demonstrate that our approach significantly improves agents' planning capabilities, providing a transformative platform toward intelligent, imaginative embodied exploration.

Crowd counting and localization involve extracting the number and distribution of crowds from images or videos using computer vision techniques. Most counting methods are based on density regression and are based on an ``intersection'' hypothesis, *i.e.*, one pixel is influenced by multiple points in the ground truth, which is inconsistent with reality since one pixel would not contain two objects. This paper proposes Proximal Mapping Loss (PML), a density regression method that eliminates this hypothesis. {PML} divides the predicted density map into multiple point-neighbor cases through the nearest neighbor, and then dynamically constructs a learning target for each sub-case via proximal mapping, leading to more robust and accurate training. {Furthermore}, PML is theoretically linked to various existing loss functions, such as Gaussian-blurred L2 loss, Bayesian loss, and the training schemes in P2PNet and DMC, demonstrating its versatility and adaptability. Experimentally, PML significantly improves the performance of crowd counting and localization, and illustrates the robustness against annotation noise. The code is available at [https://github.com/Elin24/pml](https://github.com/Elin24/pml).

Label smoothing (LS) is a popular regularisation method for training neural networks as it is effective in improving test accuracy and is simple to implement. ''Hard'' one-hot labels are ''smoothed'' by uniformly distributing probability mass to other classes, reducing overfitting. Prior work has shown that in some cases *LS can degrade selective classification (SC)* -- where the aim is to reject misclassifications using a model's uncertainty. In this work, we first demonstrate empirically across an extended range of large-scale tasks and architectures that LS *consistently* degrades SC. We then address a gap in existing knowledge, providing an *explanation* for this behaviour by analysing logit-level gradients: LS degrades the uncertainty rank ordering of correct vs incorrect predictions by regularising the max logit *more* when a prediction is likely to be correct, and *less* when it is likely to be wrong. This elucidates previously reported experimental results where strong classifiers underperform in SC. We then demonstrate the empirical effectiveness of post-hoc *logit normalisation* for recovering lost SC performance caused by LS. Furthermore, linking back to our gradient analysis, we again provide an explanation for why such normalisation is effective.

For vision-language models (VLMs), understanding the dynamic properties of objects and their interactions in 3D scenes from videos is crucial for effective reasoning about high-level temporal and action semantics. Although humans are adept at understanding these properties by constructing 3D and temporal (4D) representations of the world, current video understanding models struggle to extract these dynamic semantics, arguably because these models use cross-frame reasoning without underlying knowledge of the 3D/4D scenes. In this work, we introduce **DynSuperCLEVR**, the first video question answering dataset that focuses on language understanding of the dynamic properties of 3D objects. We concentrate on three physical concepts—*velocity*, *acceleration*, and *collisions*—within 4D scenes. We further generate three types of questions, including factual queries, future predictions, and counterfactual reasoning that involve different aspects of reasoning on these 4D dynamic properties. To further demonstrate the importance of explicit scene representations in answering these 4D dynamics questions, we propose **NS-4DPhysics**, a **N**eural-**S**ymbolic VideoQA model integrating **Physics** prior for **4D** dynamic properties with explicit scene representation of videos. Instead of answering the questions directly from the video text input, our method first estimates the 4D world states with a 3D generative model powered by a physical prior, and then uses neural symbolic reasoning to answer the questions based on the 4D world states. Our evaluation on all three types of questions in DynSuperCLEVR shows that previous video question answering models and large multimodal models struggle with questions about 4D dynamics, while our NS-4DPhysics significantly outperforms previous state-of-the-art models.

Multiple-choice question answering (MCQA) is a key competence of performant transformer language models that is tested by mainstream benchmarks. However, recent evidence shows that models can have quite a range of performance, particularly when the task format is diversified slightly (such as by shuffling answer choice order). In this work we ask: how do successful models perform formatted MCQA? We employ vocabulary projection and activation patching methods to localize key hidden states that encode relevant information for predicting the correct answer. We find that prediction of a specific answer symbol is causally attributed to a few middle layers, and specifically their multi-head self-attention mechanisms. We show that subsequent layers increase the probability of the predicted answer symbol in vocabulary space, and that this probability increase is associated with a sparse set of attention heads with unique roles. We additionally uncover differences in how different models adjust to alternative symbols. Finally, we demonstrate that a synthetic task can disentangle sources of model error to pinpoint when a model has learned formatted MCQA, and show that logit differences between answer choice tokens continue to grow over the course of training.

This research aims to comprehensively explore building a multimodal foundation model for egocentric video understanding. To achieve this goal, we work on three fronts. First, as there is a lack of QA data for egocentric video understanding, we automatically generate 7M high-quality QA samples for egocentric videos ranging from 30 seconds to one hour long in Ego4D based on human-annotated data. This is one of the largest egocentric QA datasets. Second, we contribute a challenging egocentric QA benchmark with 629 videos and 7,026 questions to evaluate the models' ability in recognizing and memorizing visual details across videos of varying lengths. We introduce a new de-biasing evaluation method to help mitigate the unavoidable language bias present in the models being evaluated. Third, we propose a specialized multimodal architecture featuring a novel ``Memory Pointer Prompting" mechanism. This design includes a global glimpse step to gain an overarching understanding of the entire video and identify key visual information, followed by a fallback step that utilizes the key visual information to generate responses. This enables the model to more effectively comprehend extended video content. With the data, benchmark, and model, we build MM-Ego, an egocentric multimodal LLM that shows powerful performance on egocentric video understanding.

Recent advancements in large language models (LLMs) have shown significant potential for automating hardware description language (HDL) code generation from high-level natural language instructions. While fine-tuning has improved LLMs' performance in hardware design tasks, prior efforts have largely focused on Verilog generation, overlooking the equally critical task of Verilog understanding. Furthermore, existing models suffer from weak alignment between natural language descriptions and Verilog code, hindering the generation of high-quality, synthesizable designs. To address these issues, we present DeepRTL, a unified representation model that excels in both Verilog understanding and generation. Based on CodeT5+, DeepRTL is fine-tuned on a comprehensive dataset that aligns Verilog code with rich, multi-level natural language descriptions. We also introduce the first benchmark for Verilog understanding and take the initiative to apply embedding similarity and GPT Score to evaluate the models' understanding capabilities. These metrics capture semantic similarity more accurately than traditional methods like BLEU and ROUGE, which are limited to surface-level n-gram overlaps. By adapting curriculum learning to train DeepRTL, we enable it to significantly outperform GPT-4 in Verilog understanding tasks, while achieving performance on par with OpenAI's o1-preview model in Verilog generation tasks.

Semi-supervised learning (SSL), exemplified by FixMatch (Sohn et al., 2020), has shown significant generalization advantages over supervised learning (SL), particularly in the context of deep neural networks (DNNs). However, it is still unclear, from a theoretical standpoint, why FixMatch-like SSL algorithms generalize better than SL on DNNs. In this work, we present the first theoretical justification for the enhanced test accuracy observed in FixMatch-like SSL applied to DNNs by taking convolutional neural networks (CNNs) on classification tasks as an example. Our theoretical analysis reveals that the semantic feature learning processes in FixMatch and SL are rather different. In particular, FixMatch learns all the discriminative features of each semantic class, while SL only randomly captures a subset of features due to the well-known lottery ticket hypothesis. Furthermore, we show that our analysis framework can be applied to other FixMatch-like SSL methods, e.g., FlexMatch, FreeMatch, Dash, and SoftMatch. Inspired by our theoretical analysis, we develop an improved variant of FixMatch, termed Semantic-Aware FixMatch (SA-FixMatch). Experimental results corroborate our theoretical findings and the enhanced generalization capability of SA-FixMatch.

Video Temporal Grounding (VTG) is a crucial capability for video understanding models and plays a vital role in downstream tasks such as video browsing and editing. To effectively handle various tasks simultaneously and enable zero-shot prediction, there is a growing trend in employing video LLMs for VTG tasks. However, current video LLM-based methods rely exclusively on natural language generation, lacking the ability to model the clear structure inherent in videos, which restricts their effectiveness in tackling VTG tasks. To address this issue, this paper first formally introduces causal event modeling framework, which represents video LLM outputs as sequences of events, and predict the current event using previous events, video inputs, and textural instructions. Each event consists of three components: timestamps, salient scores, and textual captions. We then propose a novel task-interleaved video LLM called TRACE to effectively implement the causal event modeling framework in practice. The TRACE process visual frames, timestamps, salient scores, and text as distinct tasks, employing various encoders and decoding heads for each. Task tokens are arranged in an interleaved sequence according to the causal event modeling framework's formulation. Extensive experiments on various VTG tasks and datasets demonstrate the superior performance of TRACE compared to state-of-the-art video LLMs. Our model and code are avaliable at \url{https://github.com/gyxxyg/TRACE}.