Showing papers on "Inter frame published in 2021"

Video Compressed Sensing Using a Convolutional Neural Network

Abstract: Recently, a few image compressed sensing (CS) methods based on deep learning have been developed, which achieve remarkable reconstruction quality with low computational complexity. However, these existing deep learning-based image CS methods focus on exploring intraframe correlation while ignoring interframe cues, resulting in inefficiency when directly applied to video CS. In this paper, we propose a novel video CS framework based on a convolutional neural network (dubbed VCSNet) to explore both intraframe and interframe correlations. Specifically, VCSNet divides the video sequence into multiple groups of pictures (GOPs), of which the first frame is a keyframe that is sampled at a higher sampling ratio than the other nonkeyframes. In a GOP, the block-based framewise sampling by a convolution layer is proposed, which leads to the sampling matrix being automatically optimized. In the reconstruction process, the framewise initial reconstruction by using a linear convolutional neural network is first presented, which effectively utilizes the intraframe correlation. Then, the deep reconstruction with multilevel feature compensation is proposed, which compensates the nonkeyframes with the keyframe in a multilevel feature compensation manner. Such multilevel feature compensation allows the network to better explore both intraframe and interframe correlations. Extensive experiments on six benchmark videos show that VCSNet provides better performance over state-of-the-art video CS methods and deep learning-based image CS methods in both objective and subjective reconstruction quality.

Research on Fast Pedestrian Detection Algorithm Based on Autoencoding Neural Network and AdaBoost

Abstract: In order to solve the problem of low accuracy of pedestrian detection of real traffic cameras and high missed detection rate of small target pedestrians, this paper combines autoencoding neural network and AdaBoost to construct a fast pedestrian detection algorithm Aiming at the problem that a single high-level output feature map has insufficient ability to express pedestrian features and existing methods cannot effectively select appropriate multilevel features, this paper improves the traditional AdaBoost algorithm structure, that is, the sample weight update formula and the strong classifier output formula are reset, and the two-input AdaBoost-DBN classification algorithm is proposed Moreover, in view of the problem that the fusion video is not smoothly played, this paper considers the motion information of the video object, performs pixel interpolation by motion compensation, and restores the frame rate of the original video by reconstructing the dropped interframe image Through experimental research, we can see that the algorithm constructed in this paper has a certain effect

Automatic Inter-Frame Patient Motion Correction for Dynamic Cardiac PET Using Deep Learning.

Abstract: Patient motion during dynamic PET imaging can induce errors in myocardial blood flow (MBF) estimation. Motion correction for dynamic cardiac PET is challenging because the rapid tracer kinetics of 82Rb leads to substantial tracer distribution change across different dynamic frames over time, which can cause difficulties for image registration-based motion correction, particularly for early dynamic frames. In this paper, we developed an automatic deep learning-based motion correction (DeepMC) method for dynamic cardiac PET. In this study we focused on the detection and correction of inter-frame rigid translational motion caused by voluntary body movement and pattern change of respiratory motion. A bidirectional-3D LSTM network was developed to fully utilize both local and nonlocal temporal information in the 4D dynamic image data for motion detection. The network was trained and evaluated over motion-free patient scans with simulated motion so that the motion ground-truths are available, where one million samples based on 65 patient scans were used in training, and 600 samples based on 20 patient scans were used in evaluation. The proposed method was also evaluated using additional 10 patient datasets with real motion. We demonstrated that the proposed DeepMC obtained superior performance compared to conventional registration-based methods and other convolutional neural networks (CNN), in terms of motion estimation and MBF quantification accuracy. Once trained, DeepMC is much faster than the registration-based methods and can be easily integrated into the clinical workflow. In the future work, additional investigation is needed to evaluate this approach in a clinical context with realistic patient motion.

Spatio-Temporal Deepfake Detection with Deep Neural Networks

Abstract: Deepfakes generated by generative adversarial neural networks may threaten not only individuals but also pose a public threat. In this regard, detecting video content manipulations is an urgent task, and many researchers propose various methods to solve it. Nevertheless, the problem remains. In this paper, the existing approaches are evaluated, and a new method for detecting deepfakes in videos is proposed. Considering that deepfakes are inserted into the video frame by frame, when viewing it, even with the naked eye, fluctuations and temporal distortions are noticeable, which are not taken into account by many deepfake detection algorithms that use information from a single frame to search for forgeries out of context with neighboring frames. It is proposed to analyze information from a sequence of multiple consecutive frames to detect deepfakes in video content by processing the video using the sliding window approach, taking into account not only spatial intraframe dependencies but also interframe temporal dependencies. Experiments have shown the advantage and potential for further development of the proposed approach over simple intraframe recognition.

Blind MV-based video steganalysis based on joint inter-frame and intra-frame statistics

Abstract: Despite all its irrefutable benefits, the development of steganography methods has sparked ever-increasing concerns over steganography abuse in recent decades. To prevent the inimical usage of steganography, steganalysis approaches have been introduced. Since motion vector manipulation leads to random and indirect changes in the statistics of videos, MV-based video steganography has been the center of attention in recent years. In this paper, we propose a 54-dimentional feature set exploiting spatio-temporal features of motion vectors to blindly detect MV-based stego videos. The idea behind the proposed features originates from two facts. First, there are strong dependencies among neighboring MVs due to utilizing rate-distortion optimization techniques and belonging to the same rigid object or static background. Accordingly, MV manipulation can leave important clues on the differences between each MV and the MVs belonging to the neighboring blocks. Second, a majority of MVs in original videos are locally optimal after decoding concerning the Lagrangian multiplier, notwithstanding the information loss during compression. Motion vector alteration during information embedding can affect these statistics that can be utilized for steganalysis. Experimental results have shown that our features’ performance far exceeds that of state-of-the-art steganalysis methods. This outstanding performance lies in the utilization of complementary spatio-temporal statistics affected by MV manipulation as well as feature dimensionality reduction applied to prevent overfitting. Moreover, unlike other existing MV-based steganalysis methods, our proposed features can be adjusted to various settings of the state-of-the-art video codec standards such as sub-pixel motion estimation and variable-block-size motion estimation.

Multiframe-to-Multiframe Network for Video Denoising

Abstract: Most existing studies performed video denoising by using multiple adjacent noisy frames to recover one clean frame; however, despite achieving relatively good quality for each individual frame, these approaches may result in visual flickering when the denoised frames are considered in sequence. In this paper, instead of separately restoring each clean frame, we propose a multiframe-to-multiframe (MM) denoising scheme that simultaneously recovers multiple clean frames from consecutive noisy frames. The proposed MM denoising scheme uses a training strategy that optimizes the denoised video from both the spatial and temporal dimensions, enabling better temporal consistency in the denoised video. Furthermore, we present an MM network (MMNet), which adopts a spatiotemporal convolutional architecture that considers both the interframe similarity and single-frame characteristics. Benefiting from the underlying parallel mechanism of the MM denoising scheme, MMNet achieves a highly competitive denoising efficiency. Extensive analyses and experiments demonstrate that MMNet outperforms the state-of-the-art video denoising methods, yielding temporal consistency improvements of at least 13.3% and running more than 2 times faster than the other methods.

Research on the Detection and Tracking Algorithm of Moving Object in Image Based on Computer Vision Technology

Abstract: In order to improve the video image processing technology, this paper presents a moving object detection and tracking algorithm based on computer vision technology. Firstly, the detection performance of the interframe difference method and the background difference model method is compared comprehensively from both theoretical and experimental aspects, and then the Robert edge detection operator is selected to carry out edge detection of the vehicle. The research results show that the algorithm proposed in this paper has the longest running time per frame when tracking a moving target, which is about 2.3 times that of the single frame running time of the CamShift algorithm. The algorithm has high running efficiency and can meet the requirements of real-time tracking of a foreground target. The algorithm has the highest tracking accuracy, the time consumption is reduced, and the error of the tracking frame deviating from the real position of the target is the least.

Dynamic Warping Network for Semantic Video Segmentation

Abstract: A major challenge for semantic video segmentation is how to exploit the spatiotemporal information and produce consistent results for a video sequence Many previous works utilize the precomputed optical flow to warp the feature maps across adjacent frames However, the imprecise optical flow and the warping operation without any learnable parameters may not achieve accurate feature warping and only bring a slight improvement In this paper, we propose a novel framework named Dynamic Warping Network (DWNet) to adaptively warp the interframe features for improving the accuracy of warping-based models Firstly, we design a flow refinement module (FRM) to optimize the precomputed optical flow Then, we propose a flow-guided convolution (FG-Conv) to achieve the adaptive feature warping based on the refined optical flow Furthermore, we introduce the temporal consistency loss including the feature consistency loss and prediction consistency loss to explicitly supervise the warped features instead of simple feature propagation and fusion, which guarantees the temporal consistency of video segmentation Note that our DWNet adopts extra constraints to improve the temporal consistency in the training phase, while no additional calculation and postprocessing are required during inference Extensive experiments show that our DWNet can achieve consistent improvement over various strong baselines and achieves state-of-the-art accuracy on the Cityscapes and CamVid benchmark datasets

CSANet for Video Semantic Segmentation With Inter-Frame Mutual Learning

Abstract: Video semantic segmentation aims atgenerating temporal consistent segmentation results and is still a very challenging task in the deep learning era. In this work, we improve prior approaches from two aspects. On the network architecture level, we present the cross and self-attention network (CSANet). As opposed to prior methods, CSANet not only propagates temporal features from adjacent frames, but is also designed to aggregate spatial context within the current frame, which is shown to effectively improve the consistency and robustness of the extracted deep features. On the loss function level, we further propose the inter-frame mutual learning strategy which ensures the cross-attention module to focus on semantically correlated context regions, allowing the segmentation results at different frames to be collaboratively improved. By combining the above two novel designs, we show that our proposed method is able to deliver state-of-the-art performance on the Cityscapes and CamVid benchmarks.

Coverless Video Steganography Based on Inter Frame Combination

Abstract: In most coverless image steganography methods, the number of images increases exponentially with the increase of hidden message bits, which is difficult to construct such a dataset. And several images in semantic irrelevance are usually needed to represent more secret message bits, which are easy to cause the attacker’s attention and bring some insecurity. To solve these two problems, a coverless video steganography method based on inter frame combination is proposed in this manuscript. In the proposed method, the hash sequence of a frame is generated by the CNNs and hash generator. To hide more information bits in one video, a special mapping rule is proposed. Through this mapping rule, some key frames in one video are selected. In the selected frames, one or several frames are used to represent a piece of information with equal length. To quickly index out the corresponding frames, a three-level index structure is proposed in this manuscript. Since the proposed coverless video steganography method does not embed one bit in video, it can effectively resist steganalysis algorithms. The experimental results and analysis show that the proposed method has a large capacity, good robustness and high security.

CNN-based camera motion classification using HSI color model for compressed videos

Abstract: This paper presents a novel camera motion classification framework based on modeling the compressed domain block motion vectors using the HSI color model. The input to the proposed method is the interframe block motion vectors decoded from the compressed bitstream. The block motion vector’s magnitude and orientation are estimated, followed by assigning motion vector orientation to Hue, motion vector magnitude to Saturation, and keeping Intensity at a fixed value. The HSI assignment is then converted into an RGB image followed by supervised learning utilizing a convolutional neural network to recognize eleven camera motion patterns comprising seven pure camera motion patterns and four mixed camera patterns. The proposed method’s premise is based on posing the camera motion classification problem as a color recognition task. Detailed experimental analysis that includes a comparison with state-of-the-art methods, ablation study, and robustness analysis is carried out utilizing block motion vectors obtained from H.264/AVC encoded videos. Results demonstrate accuracies of over 98 % in recognizing eleven camera patterns for the proposed method.

End-to-end Neural Video Coding Using a Compound Spatiotemporal Representation.

Abstract: Recent years have witnessed rapid advances in learnt video coding. Most algorithms have solely relied on the vector-based motion representation and resampling (e.g., optical flow based bilinear sampling) for exploiting the inter frame redundancy. In spite of the great success of adaptive kernel-based resampling (e.g., adaptive convolutions and deformable convolutions) in video prediction for uncompressed videos, integrating such approaches with rate-distortion optimization for inter frame coding has been less successful. Recognizing that each resampling solution offers unique advantages in regions with different motion and texture characteristics, we propose a hybrid motion compensation (HMC) method that adaptively combines the predictions generated by these two approaches. Specifically, we generate a compound spatiotemporal representation (CSTR) through a recurrent information aggregation (RIA) module using information from the current and multiple past frames. We further design a one-to-many decoder pipeline to generate multiple predictions from the CSTR, including vector-based resampling, adaptive kernel-based resampling, compensation mode selection maps and texture enhancements, and combines them adaptively to achieve more accurate inter prediction. Experiments show that our proposed inter coding system can provide better motion-compensated prediction and is more robust to occlusions and complex motions. Together with jointly trained intra coder and residual coder, the overall learnt hybrid coder yields the state-of-the-art coding efficiency in low-delay scenario, compared to the traditional H.264/AVC and H.265/HEVC, as well as recently published learning-based methods, in terms of both PSNR and MS-SSIM metrics.

Machine Learning-Based Multitarget Tracking of Motion in Sports Video

Abstract: In this paper, we track the motion of multiple targets in sports videos by a machine learning algorithm and study its tracking technique in depth. In terms of moving target detection, the traditional detection algorithms are analysed theoretically as well as implemented algorithmically, based on which a fusion algorithm of four interframe difference method and background averaging method is proposed for the shortcomings of interframe difference method and background difference method. The fusion algorithm uses the learning rate to update the background in real time and combines morphological processing to correct the foreground, which can effectively cope with the slow change of the background. According to the requirements of real time, accuracy, and occupying less video memory space in intelligent video surveillance systems, this paper improves the streamlined version of the algorithm. The experimental results show that the improved multitarget tracking algorithm effectively improves the Kalman filter-based algorithm to meet the real-time and accuracy requirements in intelligent video surveillance scenarios.

Silhouette 4D With Context Selection: Lossless Geometry Compression of Dynamic Point Clouds

Abstract: Recently we have proposed a coding algorithm of point cloud geometry based on a rather different approach from the popular octree representation. In our algorithm, the point cloud is decomposed in silhouettes, hence the name Silhouette Coder, and context adaptive arithmetic coding is used to exploit redundancies within the point cloud (intra frame coding), and also using a reference point cloud (inter frame coding). In this letter we build on our previous work and propose a context selection algorithm as a pre-processing stage. With this algorithm, the point cloud is first parsed testing a large number of candidate context locations. The algorithm selects a small number of these contexts that better reflect the current point cloud, and then encode it with this choice. The proposed method further improves the results of our previous coder, Silhouette 4D, by $10 \%$ , on average, on a dynamic point cloud dataset of the JPEG Pleno, and achieves bitrates competitive with some high quality lossy coders such as the MPEG G-PCC.

Optimizing the frame duration for data-driven rigid motion estimation in brain PET imaging.

Abstract: PURPOSE Data-driven rigid motion estimation for PET brain imaging is usually performed using data frames sampled at low temporal resolution to reduce the overall computation time and to provide adequate signal-to-noise ratio in the frames. In recent work it has been demonstrated that list-mode reconstructions of ultrashort frames are sufficient for motion estimation and can be performed very quickly. In this work we take the approach of using image-based registration of reconstructions of very short frames for data-driven motion estimation, and optimize a number of reconstruction and registration parameters (frame duration, MLEM iterations, image pixel size, post-smoothing filter, reference image creation, and registration metric) to ensure accurate registrations while maximizing temporal resolution and minimizing total computation time. METHODS Data from 18 F-fluorodeoxyglucose (FDG) and 18 F-florbetaben (FBB) tracer studies with varying count rates are analyzed, for PET/MR and PET/CT scanners. For framed reconstructions using various parameter combinations interframe motion is simulated and image-based registrations are performed to estimate that motion. RESULTS For FDG and FBB tracers using 4 × 105 true and scattered coincidence events per frame ensures that 95% of the registrations will be accurate to within 1 mm of the ground truth. This corresponds to a frame duration of 0.5-1 sec for typical clinical PET activity levels. Using four MLEM iterations with no subsets, a transaxial pixel size of 4 mm, a post-smoothing filter with 4-6 mm full width at half maximum, and averaging two or more frames to create the reference image provides an optimal set of parameters to produce accurate registrations while keeping the reconstruction and processing time low. CONCLUSIONS It is shown that very short frames (≤1 sec) can be used to provide accurate and quick data-driven rigid motion estimates for use in an event-by-event motion corrected reconstruction.

A progressive CNN in-loop filtering approach for inter frame coding

Abstract: Convolutional Neural Network (CNN) structures have been designed for in-loop filtering to improve video coding performance. These CNN models are usually trained through learning the correlations between the reconstructed and the original frames, which are then applied to every single reconstructed frame to improve the overall video quality. This direct model training and deployment strategy is effective for intra coding since a locally optimal model is sufficient. However, when applied to inter coding, it causes over-filtering because the intertwined reference dependencies across inter frames are not taken into consideration. To address this issue, existing methods usually resort to the Rate–Distortion Optimization (RDO) to selectively apply the CNN model, but fail to address the limitation of using a local CNN model. In this paper, we propose a progressive approach to train and incorporate the CNN-based in-loop filters to work seamlessly with video encoders. First, we develop a progressive training method to obtain the inter model. Using transfer learning, reconstructed frames using the CNN model are progressively involved back into the training of the CNN model itself, to simulate the reference dependencies in inter coding. Next, we design a frame-level model selection strategy for the high-bitrate coding where the over-filtering effect is diluted. Experimental results show that the proposed method outperforms the RDO method that utilizes only local model. Proposed approach also achieves comparable coding performance but with less computational complexity when integrating our progressive model into the RDO scheme.

Overcome the Brightness and Jitter Noises in Video Inter-Frame Tampering Detection.

Abstract: Digital video forensics plays a vital role in judicial forensics, media reports, e-commerce, finance, and public security. Although many methods have been developed, there is currently no efficient solution to real-life videos with illumination noises and jitter noises. To solve this issue, we propose a detection method that adapts to brightness and jitter for video inter-frame forgery. For videos with severe brightness changes, we relax the brightness constancy constraint and adopt intensity normalization to propose a new optical flow algorithm. For videos with large jitter noises, we introduce motion entropy to detect the jitter and extract the stable feature of texture changes fraction for double-checking. Experimental results show that, compared with previous algorithms, the proposed method is more accurate and robust for videos with significant brightness variance or videos with heavy jitter on public benchmark datasets.

Exploiting Inter-Frame Regional Correlation for Efficient Action Recognition

Abstract: Temporal feature extraction is an important issue in video-based action recognition. Optical flow is a popular method to extract temporal feature, which produces excellent performance thanks to its capacity of capturing pixel-level correlation information between consecutive frames. However, such a pixel-level correlation is extracted at the cost of high computational complexity and large storage resource. In this paper, we propose a novel temporal feature extraction method, Attentive Correlated Temporal Feature (ACTF), by exploring inter-frame correlation within a certain region. The proposed ACTF exploits both bilinear and linear correlations between successive frames on the regional level. Our method has the advantage of achieving performance comparable to or better than optical flow-based methods while avoiding the introduction of optical flow. Experimental results demonstrate our proposed method achieves the competitive performances of 96.3 % on UCF101 and 76.3 % on HMDB51 benchmark datasets.

Improved Hybrid Block-Based Motion Estimation for Inter-frame Coding

Abstract: Digital video technology has been increasingly needed in various fields, such as telecommunications, entertainment, medicine. Therefore, video compression is required. Motion estimation methods help in improving video compression efficiency by effectively removing the temporal redundancy between successive frames. Several block-based motion estimation (BME) algorithms are being suggested to reduce the coding process’s computational complexity. This paper proposes a new rapid hybrid (BME) algorithm established on the primary search point prediction and advance ending search point strategies. It combines rough adaptive search and effective local search. The coarse search introduces a new motion vector (MV) prediction technique that utilizes the macro-blocks (MBs) Spatio-temporal correlations to optimize the traditional adaptive-rood-pattern search algorithm (ARPS) and speeding up the whole process without affecting the accuracy. In the accurate local search, the cross-formed search pattern using a one-step search (OSS) block matching algorithm is employed, to estimate the actual (MV) with less computation time and further speed up the search efficiency. Exhaustive experiments are performed to demonstrate the present algorithm’s performance over the benchmark schemes concerning specific assessment criteria for results, including the peak signal-to-noise ratio (PSNR), computational complexity and computational gain. The results show that the proposed algorithm is efficient and reliable; it can always give better performance over diamond search (DS) and (ARPS). The conducted test shows an increased performance of search speed while preserving the visual quality of the motion-compensated images, and it achieves 59.76–88.03 speed improvement over (DS) and 20.98–72.06 over (ARPS) for different video sequences. Besides, the suggested method (ARP-OSS) provides the best result compared to (DS) and (ARPS) in terms of time complexity for analyzing all the video samples.

An Algorithm for Motion Estimation Based on the Interframe Difference Detection Function Model

Abstract: In this paper, we simulate the estimation of motion through an interframe difference detection function model and investigate the spatial-temporal context information correlation filtering target tracking algorithm, which is complex and computationally intensive. The basic theory of spatiotemporal context information and correlation filtering is studied to construct a fast target tracking method. The different computational schemes are designed for the flow of multiframe target detection from background removal to noise reduction, to single-frame detection, and finally to multiframe detection, respectively. This enables the ground-based telescope to effectively detect spatial targets in dense stellar backgrounds in both modes. The method is validated by simulations and experiments and can meet the requirements of real projects. The interframe bit attitude estimation is optimized by using the beam-parity method to reduce the interframe estimation noise; a global optimization strategy based on the bit attitude map is used in the back end to reduce the system computation amount and make the global bit attitude estimation more accurate; a loop detection based on the word pocket model is added to the system to reduce the cumulative error.

Adaptive GOP Size Decision for Multi-Pass Video Coding Based on Hidden Markov Model

Abstract: Multi-pass coding is a widely utilized technique to improve the compression efficiency in video coding, where frame statistics are collected from the previous passes and then analyzed to provide better encoder decisions, such as rate control parameters, prediction mode selection, motion estimation, etc. In this paper, a novel method to determine the size of each group of picture (GOP) using the multi-pass information is presented. In particular, we propose to categorize frames into regions with different natures, including stationary, high-variance, blending, and scene cut, through analyzing the frame statistics generated from the previous passes using a hidden Markov model. The GOP size is then determined based on the region types and the inter frame correlations. It is experimentally shown that the proposed adaptive GOP size decision provides considerable coding performance improvements over conventional fixed GOP length.

Gated Fusion Network for SAO Filter and Inter Frame Prediction in Versatile Video Coding

Abstract: To achieve higher coding efficiency, Versatile Video Coding (VVC) includes several novel components, but at the expense of increasing decoder computational complexity. These technologies at a low bit rate often create contouring and ringing effects on the reconstructed frames and introduce various blocking artifacts at block boundaries. To suppress those visual artifacts, the VVC framework supports four post-processing filter operations. The interoperation of these filters introduces extra signaling bits and eventually becomes overhead at higher resolution video processing. In this paper, a novel deep learning-based model is proposed for sample adaptive offset (SAO) nonlinear filtering operation and substantiated the merits of intra-inter frame quality enhancement. We introduced a variable filter size multi-scale CNN (MSCNN) to improve the denoising operation and incorporated strided deconvolution for further computation improvement. We demonstrated that our deconvolution model can effectively be trained by leveraging the high-frequency edge features learned in a parallel fashion using feature fusion and residual learning. The simulation results demonstrate that the proposed method outperforms the baseline VVC method in BD-BR, BD-PSNR measurements and achieves an average of 3.762 % bit rate saving on the standard video test sequences.

Video Instance Segmentation using Inter-Frame Communication Transformers

Abstract: We propose a novel end-to-end solution for video instance segmentation (VIS) based on transformers. Recently, the per-clip pipeline shows superior performance over per-frame methods leveraging richer information from multiple frames. However, previous per-clip models require heavy computation and memory usage to achieve frame-to-frame communications, limiting practicality. In this work, we propose Inter-frame Communication Transformers (IFC), which significantly reduces the overhead for information-passing between frames by efficiently encoding the context within the input clip. Specifically, we propose to utilize concise memory tokens as a mean of conveying information as well as summarizing each frame scene. The features of each frame are enriched and correlated with other frames through exchange of information between the precisely encoded memory tokens. We validate our method on the latest benchmark sets and achieved the state-of-the-art performance (AP 44.6 on YouTube-VIS 2019 val set using the offline inference) while having a considerably fast runtime (89.4 FPS). Our method can also be applied to near-online inference for processing a video in real-time with only a small delay. The code will be made available.

IoTorch: Reliable LED-to-Camera Communication Against Inter-Frame Gaps and Frame Drops

Abstract: LED-to-camera communication for smartphones promises to enable low-cost, small-size, and intuitive data access to Internet-of-things (IoT) devices. However, for fast and reliable communication under the rolling shutter effect, the packet losses caused by inter-frame gaps and frame drops must be dealt with. In this paper, we introduce LED-to-camera communication between a commercial smartphone and an IoT device with LED flashlight, where the user can intuitively acquire the desired data transmitted from the IoT device to the smartphone. Specifically, we present IoTorch, a fast and reliable LED-to-camera communication that efficiently prevents the packet losses. IoTorch consists of two core mechanisms: 1) a minimum-repetition one-way reliable transmission focusing on the periodicity of inter-frame gaps and 2) an acknowledgement mechanism to overcome frame drops by using a smartphone's built-in flash focusing on its delay characteristics. Additionally, we propose an optimization method to increase the throughput up to 2.92 kbps (i.e., 2.43 times faster than that of the state-of-the-art method that overcomes the inter-frame gaps). Furthermore, we realize a remote wake-up function by using the smartphone's flash as the data transmission trigger. To demonstrate the benefit of IoTorch, we present a sensor data viewer using Arduino and a smartphone console system on TelosB.