Text-based video editing has recently attracted considerable interest in changing the style or replacing the objects with a similar structure. Beyond this, we demonstrate that properties such as shape, size, location, motion, etc., can also be edited in videos. Our key insight is that the keyframe's transformations of the specific internal feature (e.g., edge maps of objects or human pose), can easily propagate to other frames to provide generation guidance. We thus propose MagicStick, a controllable video editing method that edits the video properties by utilizing the transformation on the extracted internal control signals. In detail, to keep the appearance, we inflate both the pre-trained image diffusion model and ControlNet to the temporal dimension and train low-rank adaptions (LoRA) layers to fit the specific scenes. Then, in editing, we perform an inversion and editing framework. Differently, finetuned ControlNet is introduced in both inversion and generation for attention guidance with the proposed attention remix between the spatial attention maps of inversion and editing. Yet succinct, our method is the first method to show the ability of video property editing from the pre-trained text-to-image model. We present experiments on numerous examples within our unified framework. We also compare with shape-aware text-based editing and handcrafted motion video generation, demonstrating our superior temporal consistency and editing capability than previous works.

Generative diffusion models deliver remarkable synthesis quality but remain impractical for resource-limited Internet of Things (IoT) devices due to their substantial computational and memory demands. To bridge this critical gap, we present a comprehensive, multi-stage optimization framework that systematically reduces model size while meticulously preserving generative fidelity. The framework integrates an efficient backbone architecture designed for inherent lightness, a sensitivity-guided fine-grained pruning strategy that strategically removes redundant parameters to achieve high sparsity, and a novel distribution-aware quantization algorithm based on Gaussian Mixture Models (GMMs) to compress weights and activations with minimal quality degradation. Extensive validation across multiple diffusion architectures (DDPM, DDIM, SGM) and diverse datasets demonstrates the framework’s strong generalizability, achieving up to 79% model sparsity while preserving generative fidelity. To showcase practical utility, we demonstrate that our framework produces a compressed model compatible with standard mobile deployment toolchains, realizing a significant reduction in the on-device memory footprint required for inference. This work offers a robust and generalizable methodology for enabling advanced generative AI on a wide spectrum of edge and IoT platforms. Code is available at: https://github.com/mitchell-cheng/compress_diffusion.

Video class-incremental learning aims to recognize new actions while restricting the catastrophic forgetting of old ones, whose representative samples can only be saved in limited memory. Semantically variable subactions are susceptible to class confusion due to data imbalance. While existing methods address the problem by estimating and distilling the spatio-temporal knowledge, we further explores that the refinement of hierarchical correlations is crucial for the alignment of spatio-temporal features. To enhance the adaptability on evolved actions, we proposes a hierarchical aggregation strategy, in which hierarchical matching matrices are combined and jointly optimized to selectively store and retrieve relevant features from previous tasks. Meanwhile, a correlation refinement mechanism is presented to reinforce the bias on informative exemplars according to online hypercorrelation distribution. Experimental results demonstrate the effectiveness of the proposed method on three standard video class-incremental learning benchmarks, outperforming state-of-the-art methods. Code is available at: https://github.com/Lsen991031/HCE

Research has shown that agile language Chisel and related agile design methodology is promising to sustain the scaling computing performance in a more efficient way. Design For Test, as an economical and effective method for chip production testing, must be integrated into the agile development system to meet the requirements of mass production. Due to the lack of support for the traditional EDA toolchain, chip development through Chisel has not been widely accepted in the industry. This paper is based on the research of XiangShan, an open-source project for RISC-V high-performance processor. XiangShan establishes a structured DFT(degisn for test) development approach and Chisel-based DFT agile design flow. XiangShan develops a flexible chisel-based XS-shared bus mbist interface to improve design PPA and proposes the MarchSLD algorithm to enhance defect detection in the FinFet process node. XiangShan’s DFT development Approach is moving towards hierarchical, flexible, and reliable in two generations of processors. Experimental results show that the optimized Chisel-based DFT design flow can support agile development requirements and achieve industry-competitive performance.

Deterministic models dominate the field of talking facial land-mark generation by directly mapping speech signals to a certain lip-sync facial landmark sequence, which often suffer from regression to the mean face. In contrast, probability generative models are more beneficial to handle complex data space and generate diverse samples. In this work, we pro-pose a flow-based probabilistic network named TalkingFlow to generate natural talking facial landmark with head movements from speech data. It is implemented by a weighted multi-scale architecture to improve model representation capability and a conditional Temporal Convolutional Network module to fuse speech data. Extensive experiments results show that it can effectively generate diverse and natural facial landmark from speech data. All code will be made publicly available online.

We propose a tester-based symbolic algorithm to incrementally generate tree-like witnesses (the dual concept of counterexamples) for branching-time logic CTL*. The CTL* witnesses are state transition directed graphs of tree-like structure in which all strongly connected components are cycles, and all branches are lasso paths with shortest prefixes. The states and transitions in these graphs may be respectively attached with satisfied state and path subformulas as annotations for intelligibility. The correctness of the lasso path generation algorithm is proved. We have developed an ordered binary decision diagrams based symbolic model checker MCTK2 for CTL* in which the proposed witness generation algorithm has been implemented. To make the generated witnesses graphs more succinct, several optimization strategies are applied to reduce the number of annotations and avoid constructing branching paths as much as possible. The interactive visualization interface of MCTK2 enables users to conveniently inspect counterexamples and expand branches of interest such that they can incrementally analyze larger and more complex counterexamples.

Generating high-fidelity talking head video by fitting with the input audio sequence is a challenging problem that receives considerable attentions recently. In this paper, we address this problem with the aid of neural scene representation networks. Our method is completely different from existing methods that rely on intermediate representations like 2D landmarks or 3D face models to bridge the gap between audio input and video output. Specifically, the feature of input audio signal is directly fed into a conditional implicit function to generate a dynamic neural radiance field, from which a high-fidelity talking-head video corresponding to the audio signal is synthesized using volume rendering. Another advantage of our framework is that not only the head (with hair) region is synthesized as previous methods did, but also the upper body is generated via two individual neural radiance fields. Experimental results demonstrate that our novel framework can (1) produce high-fidelity and natural results, and (2) support free adjustment of audio signals, viewing directions, and background images. Code is available at https://github.com/YudongGuo/AD-NeRF.

With the widespread adoption of location-based services, users are increasingly demanding high- precision indoor localization. However, the deployment of localization network elements, i.e., localization base stations (LBS), mostly depends on experiences which usually leads to an extreme deployment cost. We therefore develop an indoor localization simulation platform, which can obtain the error distributions of the localization system under different LBS deployments, and also provide a near-optimal LBS deployment in practice.

Recently, character model based on integrated convolutional neural network (CNN) and deep bidirectional long short-term memory (DBLSTM) has been demonstrated to be effective for online handwritten Chinese text recognition (HCTR). However, the reported CNN-DBLSTM topologies are too complex to be practically useful. In this paper, we propose a compact CNN-DBLSTM which has small footprint and low computation cost yet be able to accommodate multiple receptive fields for CNN-based feature extraction. By using the training set of a popular benchmark database, namely CASIA-OLHWDB, we trained a compact CNN-DBLSTM by a connectionist temporal classification (CTC) criterion with a multi-step training strategy. Combined this character model with a character trigram language model, our online HCTR system with a WFSTbased decoder has achieved state-of-the-art performance on both CASIA and ICDAR-2013 Chinese handwriting recognition competition test sets.

The accuracy of a typical state-of-the-art optical character recognition (OCR) system benefits greatly from using a language model (LM). However, a conventional LM has a limited vocabulary, resulting in out-of-vocabulary (OOV) words that cannot be recognized by the OCR system. In this paper, we present an open vocabulary OCR system based on a hybrid LM. The vocabulary of the hybrid LM consists of both words and subwords. OOV words can be generated by combinations of subwords. A refined hybrid LM training scheme is applied by interpolating a standard hybrid LM, a word-based LM and a subword-based LM. An efficient word combination method is performed by modeling optional space symbols in a decoding network. The overall system deals with OOV words in a general, data-driven and language-independent way. We conduct experiments on an English handwriting OCR task. Evaluations on three testing sets demonstrate that the OCR system with the proposed method achieves a word error rate of 33.4% on an OOV-only testing set, yet without degrading the recognition accuracies on the other two testing sets mainly consisting of in-vocabulary words.

Recently, character model based on integrated convolutional neural network (CNN) and deep bidirectional long short-term memory (DBLSTM) has achieved excellent performance for offline handwriting recognition (HWR). To deploy CNN-DBLSTM model in products, it is necessary to reduce the footprint and runtime latency as much as possible. In this paper, we study two methods to compress the CNN part: (1) Use Tucker decomposition to decompose pre-trained weights with low-rank approximation, followed by fine-tuning; (2) Use grouped convolution to construct sparse connections in channel domain. Experiments have been conducted on a large-scale offline English HWR task to compare the effectiveness of the above two techniques. Our results show that using Tucker decomposition alone offers a good solution to building a compact CNN-DBLSTM model which can reduce significantly both the footprint and latency yet without degrading recognition accuracy.

We study two sequence discriminative training criteria, i.e., Lattice-Free Maximum Mutual Information (LFMMI) and Connectionist Temporal Classification (CTC), for end-to-end training of Deep Bidirectional Long Short-Term Memory (DBLSTM) based character models of two offline English handwriting recognition systems with an input feature vector sequence extracted by Principal Component Analysis (PCA) and Convolutional Neural Network (CNN), respectively. We observe that refining CTC-trained PCA-DBLSTM model with an interpolated CTC and LFMMI objective function ("CTC+LFMMI") for several additional iterations achieves a relative Word Error Rate (WER) reduction of 24.6% and 13.9% on the public IAM test set and an in-house E2E test set, respectively. For a much better CTC-trained CNN-DBLSTM system, the proposed "CTC+LFMMI" method achieves a relative WER reduction of 19.6% and 8.3% on the above two test sets, respectively.