Min Sun

Fu-En Wang, Yu-Hsuan Yeh, Yi-Hsuan Tsai et al.

Due to the rise of spherical cameras, monocular 360 depth estimation becomes an important technique for many applications (e.g., autonomous systems). Thus, state-of-the-art frameworks for monocular 360 depth estimation such as bi-projection fusion in BiFuse are proposed. To train such a framework, a large number of panoramas along with the corresponding depth ground truths captured by laser sensors are required, which highly increases the cost of data collection. Moreover, since such a data collection procedure is time-consuming, the scalability of extending these methods to different scenes becomes a challenge. To this end, self-training a network for monocular depth estimation from 360 videos is one way to alleviate this issue. However, there are no existing frameworks that incorporate bi-projection fusion into the self-training scheme, which highly limits the self-supervised performance since bi-projection fusion can leverage information from different projection types. In this paper, we propose BiFuse++ to explore the combination of bi-projection fusion and the self-training scenario. To be specific, we propose a new fusion module and Contrast-Aware Photometric Loss to improve the performance of BiFuse and increase the stability of self-training on real-world videos. We conduct both supervised and self-supervised experiments on benchmark datasets and achieve state-of-the-art performance.

Bolivar Solarte, Chin-Hsuan Wu, Jin-Cheng Jhang et al.

In this paper, we introduce a novel geometry-aware self-training framework for room layout estimation models on unseen scenes with unlabeled data. Our approach utilizes a ray-casting formulation to aggregate multiple estimates from different viewing positions, enabling the computation of reliable pseudo-labels for self-training. In particular, our ray-casting approach enforces multi-view consistency along all ray directions and prioritizes spatial proximity to the camera view for geometry reasoning. As a result, our geometry-aware pseudo-labels effectively handle complex room geometries and occluded walls without relying on assumptions such as Manhattan World or planar room walls. Evaluation on publicly available datasets, including synthetic and real-world scenarios, demonstrates significant improvements in current state-of-the-art layout models without using any human annotation.

YuanFu Yang, Min Sun

Semiconductor manufacturing is on the cusp of a revolution: the Internet of Things (IoT). With IoT we can connect all the equipment and feed information back to the factory so that quality issues can be detected. In this situation, more and more edge devices are used in wafer inspection equipment. This edge device must have the ability to quickly detect defects. Therefore, how to develop a high-efficiency architecture for automatic defect classification to be suitable for edge devices is the primary task. In this paper, we present a novel architecture that can perform defect classification in a more efficient way. The first function is self-proliferation, using a series of linear transformations to generate more feature maps at a cheaper cost. The second function is self-attention, capturing the long-range dependencies of feature map by the channel-wise and spatial-wise attention mechanism. We named this method as self-proliferation-and-attention neural network. This method has been successfully applied to various defect pattern classification tasks. Compared with other latest methods, SP&A-Net has higher accuracy and lower computation cost in many defect inspection tasks.

Tao Tu, Ming-Feng Li, Chieh Hubert Lin et al.

Articulated 3D reconstruction has valuable applications in various domains, yet it remains costly and demands intensive work from domain experts. Recent advancements in template-free learning methods show promising results with monocular videos. Nevertheless, these approaches necessitate a comprehensive coverage of all viewpoints of the subject in the input video, thus limiting their applicability to casually captured videos from online sources. In this work, we study articulated 3D shape reconstruction from a single and casually captured internet video, where the subject's view coverage is incomplete. We propose DreaMo that jointly performs shape reconstruction while solving the challenging low-coverage regions with view-conditioned diffusion prior and several tailored regularizations. In addition, we introduce a skeleton generation strategy to create human-interpretable skeletons from the learned neural bones and skinning weights. We conduct our study on a self-collected internet video collection characterized by incomplete view coverage. DreaMo shows promising quality in novel-view rendering, detailed articulated shape reconstruction, and skeleton generation. Extensive qualitative and quantitative studies validate the efficacy of each proposed component, and show existing methods are unable to solve correct geometry due to the incomplete view coverage.

Jin-Cheng Jhang, Tao Tu, Fu-En Wang et al.

The field of indoor monocular 3D object detection is gaining significant attention, fueled by the increasing demand in VR/AR and robotic applications. However, its advancement is impeded by the limited availability and diversity of 3D training data, owing to the labor-intensive nature of 3D data collection and annotation processes. In this paper, we present V-MIND (Versatile Monocular INdoor Detector), which enhances the performance of indoor 3D detectors across a diverse set of object classes by harnessing publicly available large-scale 2D datasets. By leveraging well-established monocular depth estimation techniques and camera intrinsic predictors, we can generate 3D training data by converting large-scale 2D images into 3D point clouds and subsequently deriving pseudo 3D bounding boxes. To mitigate distance errors inherent in the converted point clouds, we introduce a novel 3D self-calibration loss for refining the pseudo 3D bounding boxes during training. Additionally, we propose a novel ambiguity loss to address the ambiguity that arises when introducing new classes from 2D datasets. Finally, through joint training with existing 3D datasets and pseudo 3D bounding boxes derived from 2D datasets, V-MIND achieves state-of-the-art object detection performance across a wide range of classes on the Omni3D indoor dataset.

Tzu-Chun Chien, Chieh-Kai Lin, Shiang-Feng Tsai et al.

Recent Multimodal Large Language Models (MLLMs) have demonstrated strong performance in visual grounding, establishing themselves as a general interface for various vision-language applications. This progress has driven the development of token pruning methods to mitigate the high computational costs associated with processing numerous visual tokens. However, we observe that pruning significantly weakens the model's grounding ability, leading to incorrect predictions and drastic performance degradation. In Referring Expression Comprehension (REC), for instance, pruning causes the accuracy of LLaVA on the RefCOCO validation set to drop from 56.14% to 15.34%. Our analysis identifies misaligned position IDs after pruning as the primary cause of this degradation, as both the order and value of these IDs are crucial for maintaining performance in grounding tasks. To address this issue, we propose Grounding-Aware Token Pruning (GAP), a simple yet effective adjustment to position IDs that recovers REC accuracy back to 51.42%, which is 90% of the original performance in the without pruning setting, all while requiring no additional training, memory, or computational overhead. Applied to models such as Shikra, MiniGPTv2, and the LLaVA series, our method consistently improves performance across various token pruning strategies.

Hritam Basak, Hadi Tabatabaee, Shreekant Gayaka et al.

3D object generation from a single image involves estimating the full 3D geometry and texture of unseen views from an unposed RGB image captured in the wild. Accurately reconstructing an object's complete 3D structure and texture has numerous applications in real-world scenarios, including robotic manipulation, grasping, 3D scene understanding, and AR/VR. Recent advancements in 3D object generation have introduced techniques that reconstruct an object's 3D shape and texture by optimizing the efficient representation of Gaussian Splatting, guided by pre-trained 2D or 3D diffusion models. However, a notable disparity exists between the training datasets of these models, leading to distinct differences in their outputs. While 2D models generate highly detailed visuals, they lack cross-view consistency in geometry and texture. In contrast, 3D models ensure consistency across different views but often result in overly smooth textures. We propose bridging the gap between 2D and 3D diffusion models to address this limitation by integrating a two-stage frequency-based distillation loss with Gaussian Splatting. Specifically, we leverage geometric priors in the low-frequency spectrum from a 3D diffusion model to maintain consistent geometry and use a 2D diffusion model to refine the fidelity and texture in the high-frequency spectrum of the generated 3D structure, resulting in more detailed and fine-grained outcomes. Our approach enhances geometric consistency and visual quality, outperforming the current SOTA. Additionally, we demonstrate the easy adaptability of our method for efficient object pose estimation and tracking.

Sanghun Jung, Jingjing Zheng, Ke Zhang et al.

Unlike closed-vocabulary 3D instance segmentation that is often trained end-to-end, open-vocabulary 3D instance segmentation (OV-3DIS) often leverages vision-language models (VLMs) to generate 3D instance proposals and classify them. While various concepts have been proposed from existing research, we observe that these individual concepts are not mutually exclusive but complementary. In this paper, we propose a new state-of-the-art solution for OV-3DIS by carefully designing a recipe to combine the concepts together and refining them to address key challenges. Our solution follows the two-stage scheme: 3D proposal generation and instance classification. We employ robust 3D tracking-based proposal aggregation to generate 3D proposals and remove overlapped or partial proposals by iterative merging/removal. For the classification stage, we replace the standard CLIP model with Alpha-CLIP, which incorporates object masks as an alpha channel to reduce background noise and obtain object-centric representation. Additionally, we introduce the standardized maximum similarity (SMS) score to normalize text-to-proposal similarity, effectively filtering out false positives and boosting precision. Our framework achieves state-of-the-art performance on ScanNet200 and S3DIS across all AP and AR metrics, even surpassing an end-to-end closed-vocabulary method.

Wei-Cheng Tseng, Po-Han Chi, Jia-Hua Wu et al.

The capability of accurate prediction of protein functions and properties is essential in the biotechnology industry, e.g. drug development and artificial protein synthesis, etc. The main challenges of protein function prediction are the large label space and the lack of labeled training data. Our method leverages unsupervised sequence embedding and the success of deep convolutional neural network to overcome these challenges. In contrast, most of the existing methods delete the rare protein functions to reduce the label space. Furthermore, some existing methods require additional bio-information (e.g., the 3-dimensional structure of the proteins) which is difficult to be determined in biochemical experiments. Our proposed method significantly outperforms the other methods on the publicly available benchmark using only protein sequences as input. This allows the process of identifying protein functions to be sped up.

Chi-Wei Hsiao, Cheng Sun, Hwann-Tzong Chen et al.

We present a novel pyramidal output representation to ensure parsimony with our "specialize and fuse" process for semantic segmentation. A pyramidal "output" representation consists of coarse-to-fine levels, where each level is "specialize" in a different class distribution (e.g., more stuff than things classes at coarser levels). Two types of pyramidal outputs (i.e., unity and semantic pyramid) are "fused" into the final semantic output, where the unity pyramid indicates unity-cells (i.e., all pixels in such cell share the same semantic label). The process ensures parsimony by predicting a relatively small number of labels for unity-cells (e.g., a large cell of grass) to build the final semantic output. In addition to the "output" representation, we design a coarse-to-fine contextual module to aggregate the "features" representation from different levels. We validate the effectiveness of each key module in our method through comprehensive ablation studies. Finally, our approach achieves state-of-the-art performance on three widely-used semantic segmentation datasets -- ADE20K, COCO-Stuff, and Pascal-Context.

An-Chieh Cheng, Xueting Li, Min Sun et al.

We propose a canonical point autoencoder (CPAE) that predicts dense correspondences between 3D shapes of the same category. The autoencoder performs two key functions: (a) encoding an arbitrarily ordered point cloud to a canonical primitive, e.g., a sphere, and (b) decoding the primitive back to the original input instance shape. As being placed in the bottleneck, this primitive plays a key role to map all the unordered point clouds on the canonical surface and to be reconstructed in an ordered fashion. Once trained, points from different shape instances that are mapped to the same locations on the primitive surface are determined to be a pair of correspondence. Our method does not require any form of annotation or self-supervised part segmentation network and can handle unaligned input point clouds. Experimental results on 3D semantic keypoint transfer and part segmentation transfer show that our model performs favorably against state-of-the-art correspondence learning methods.

Bolivar Solarte, Chin-Hsuan Wu, Kuan-Wei Lu et al.

This paper presents a novel preconditioning strategy for the classic 8-point algorithm (8-PA) for estimating an essential matrix from 360-FoV images (i.e., equirectangular images) in spherical projection. To alleviate the effect of uneven key-feature distributions and outlier correspondences, which can potentially decrease the accuracy of an essential matrix, our method optimizes a non-rigid transformation to deform a spherical camera into a new spatial domain, defining a new constraint and a more robust and accurate solution for an essential matrix. Through several experiments using random synthetic points, 360-FoV, and fish-eye images, we demonstrate that our normalization can increase the camera pose accuracy by about 20% without significantly overhead the computation time. In addition, we present further benefits of our method through both a constant weighted least-square optimization that improves further the well known Gold Standard Method (GSM) (i.e., the non-linear optimization by using epipolar errors); and a relaxation of the number of RANSAC iterations, both showing that our normalization outcomes a more reliable, robust, and accurate solution.

Wei-Cheng Tseng, Jin-Siang Lin, Yao-Min Feng et al.

Humans can master a new task within a few trials by drawing upon skills acquired through prior experience. To mimic this capability, hierarchical models combining primitive policies learned from prior tasks have been proposed. However, these methods fall short comparing to the human's range of transferability. We propose a method, which leverages the hierarchical structure to train the combination function and adapt the set of diverse primitive polices alternatively, to efficiently produce a range of complex behaviors on challenging new tasks. We also design two regularization terms to improve the diversity and utilization rate of the primitives in the pre-training phase. We demonstrate that our method outperforms other recent policy transfer methods by combining and adapting these reusable primitives in tasks with continuous action space. The experiment results further show that our approach provides a broader transferring range. The ablation study also shows the regularization terms are critical for long range policy transfer. Finally, we show that our method consistently outperforms other methods when the quality of the primitives varies.

Chun-Hung Chao, Hsien-Tzu Cheng, Tsung-Ying Ho et al.

Gross tumor volume (GTV) delineation on tomography medical imaging is crucial for radiotherapy planning and cancer diagnosis. Convolutional neural networks (CNNs) has been predominated on automatic 3D medical segmentation tasks, including contouring the radiotherapy target given 3D CT volume. While CNNs may provide feasible outcome, in clinical scenario, double-check and prediction refinement by experts is still necessary because of CNNs' inconsistent performance on unexpected patient cases. To provide experts an efficient way to modify the CNN predictions without retrain the model, we propose 3D-fused context propagation, which propagates any edited slice to the whole 3D volume. By considering the high-level feature maps, the radiation oncologists would only required to edit few slices to guide the correction and refine the whole prediction volume. Specifically, we leverage the backpropagation for activation technique to convey the user editing information backwardly to the latent space and generate new prediction based on the updated and original feature. During the interaction, our proposed approach reuses the extant extracted features and does not alter the existing 3D CNN model architectures, avoiding the perturbation on other predictions. The proposed method is evaluated on two published radiotherapy target contouring datasets of nasopharyngeal and esophageal cancer. The experimental results demonstrate that our proposed method is able to further effectively improve the existing segmentation prediction from different model architectures given oncologists' interactive inputs.

Chun-Hung Chao, Zhuotun Zhu, Dazhou Guo et al.

Determining the spread of GTV$_{LN}$ is essential in defining the respective resection or irradiating regions for the downstream workflows of surgical resection and radiotherapy for many cancers. Different from the more common enlarged lymph node (LN), GTV$_{LN}$ also includes smaller ones if associated with high positron emission tomography signals and/or any metastasis signs in CT. This is a daunting task. In this work, we propose a unified LN appearance and inter-LN relationship learning framework to detect the true GTV$_{LN}$. This is motivated by the prior clinical knowledge that LNs form a connected lymphatic system, and the spread of cancer cells among LNs often follows certain pathways. Specifically, we first utilize a 3D convolutional neural network with ROI-pooling to extract the GTV$_{LN}$'s instance-wise appearance features. Next, we introduce a graph neural network to further model the inter-LN relationships where the global LN-tumor spatial priors are included in the learning process. This leads to an end-to-end trainable network to detect by classifying GTV$_{LN}$. We operate our model on a set of GTV$_{LN}$ candidates generated by a preliminary 1st-stage method, which has a sensitivity of $>85\%$ at the cost of high false positive (FP) ($>15$ FPs per patient). We validate our approach on a radiotherapy dataset with 142 paired PET/RTCT scans containing the chest and upper abdominal body parts. The proposed method significantly improves over the state-of-the-art (SOTA) LN classification method by $5.5\%$ and $13.1\%$ in F1 score and the averaged sensitivity value at $2, 3, 4, 6$ FPs per patient, respectively.

Yen-Chi Cheng, Hsin-Ying Lee, Min Sun et al.

Flexible user controls are desirable for content creation and image editing. A semantic map is commonly used intermediate representation for conditional image generation. Compared to the operation on raw RGB pixels, the semantic map enables simpler user modification. In this work, we specifically target at generating semantic maps given a label-set consisting of desired categories. The proposed framework, SegVAE, synthesizes semantic maps in an iterative manner using conditional variational autoencoder. Quantitative and qualitative experiments demonstrate that the proposed model can generate realistic and diverse semantic maps. We also apply an off-the-shelf image-to-image translation model to generate realistic RGB images to better understand the quality of the synthesized semantic maps. Furthermore, we showcase several real-world image-editing applications including object removal, object insertion, and object replacement.

Fu-En Wang, Yu-Hsuan Yeh, Min Sun et al.

Inferring the information of 3D layout from a single equirectangular panorama is crucial for numerous applications of virtual reality or robotics (e.g., scene understanding and navigation). To achieve this, several datasets are collected for the task of 360 layout estimation. To facilitate the learning algorithms for autonomous systems in indoor scenarios, we consider the Matterport3D dataset with their originally provided depth map ground truths and further release our annotations for layout ground truths from a subset of Matterport3D. As Matterport3D contains accurate depth ground truths from time-of-flight (ToF) sensors, our dataset provides both the layout and depth information, which enables the opportunity to explore the environment by integrating both cues.

Shih-Han Chou, Wei-Lun Chao, Wei-Sheng Lai et al.

In this work, we introduce VQA 360, a novel task of visual question answering on 360 images. Unlike a normal field-of-view image, a 360 image captures the entire visual content around the optical center of a camera, demanding more sophisticated spatial understanding and reasoning. To address this problem, we collect the first VQA 360 dataset, containing around 17,000 real-world image-question-answer triplets for a variety of question types. We then study two different VQA models on VQA 360, including one conventional model that takes an equirectangular image (with intrinsic distortion) as input and one dedicated model that first projects a 360 image onto cubemaps and subsequently aggregates the information from multiple spatial resolutions. We demonstrate that the cubemap-based model with multi-level fusion and attention diffusion performs favorably against other variants and the equirectangular-based models. Nevertheless, the gap between the humans' and machines' performance reveals the need for more advanced VQA 360 algorithms. We, therefore, expect our dataset and studies to serve as the benchmark for future development in this challenging task. Dataset, code, and pre-trained models are available online.

Shih-Han Chou, Cheng Sun, Wen-Yen Chang et al.

While there are several widely used object detection datasets, current computer vision algorithms are still limited in conventional images. Such images narrow our vision in a restricted region. On the other hand, 360° images provide a thorough sight. In this paper, our goal is to provide a standard dataset to facilitate the vision and machine learning communities in 360° domain. To facilitate the research, we present a real-world 360° panoramic object detection dataset, 360-Indoor, which is a new benchmark for visual object detection and class recognition in 360° indoor images. It is achieved by gathering images of complex indoor scenes containing common objects and the intensive annotated bounding field-of-view. In addition, 360-Indoor has several distinct properties: (1) the largest category number (37 labels in total). (2) the most complete annotations on average (27 bounding boxes per image). The selected 37 objects are all common in indoor scene. With around 3k images and 90k labels in total, 360-Indoor achieves the largest dataset for detection in 360° images. In the end, extensive experiments on the state-of-the-art methods for both classification and detection are provided. We will release this dataset in the near future.

Chi-Wei Hsiao, Cheng Sun, Min Sun et al.

This paper proposes a new approach, Flat2Layout, for estimating general indoor room layout from a single-view RGB image whereas existing methods can only produce layout topologies captured from the box-shaped room. The proposed flat representation encodes the layout information into row vectors which are treated as the training target of the deep model. A dynamic programming based postprocessing is employed to decode the estimated flat output from the deep model into the final room layout. Flat2Layout achieves state-of-the-art performance on existing room layout benchmark. This paper also constructs a benchmark for validating the performance on general layout topologies, where Flat2Layout achieves good performance on general room types. Flat2Layout is applicable on more scenario for layout estimation and would have an impact on applications of Scene Modeling, Robotics, and Augmented Reality.

Tsun-Hsuan Wang, Yen-Chi Cheng, Chieh Hubert Lin et al.

While image manipulation achieves tremendous breakthroughs (e.g., generating realistic faces) in recent years, video generation is much less explored and harder to control, which limits its applications in the real world. For instance, video editing requires temporal coherence across multiple clips and thus poses both start and end constraints within a video sequence. We introduce point-to-point video generation that controls the generation process with two control points: the targeted start- and end-frames. The task is challenging since the model not only generates a smooth transition of frames, but also plans ahead to ensure that the generated end-frame conforms to the targeted end-frame for videos of various length. We propose to maximize the modified variational lower bound of conditional data likelihood under a skip-frame training strategy. Our model can generate sequences such that their end-frame is consistent with the targeted end-frame without loss of quality and diversity. Extensive experiments are conducted on Stochastic Moving MNIST, Weizmann Human Action, and Human3.6M to evaluate the effectiveness of the proposed method. We demonstrate our method under a series of scenarios (e.g., dynamic length generation) and the qualitative results showcase the potential and merits of point-to-point generation. For project page, see https://zswang666.github.io/P2PVG-Project-Page/

Fang-I Hsiao, Jui-Hsuan Kuo, Min Sun

We propose a novel approach to train a multi-modal policy from mixed demonstrations without their behavior labels. We develop a method to discover the latent factors of variation in the demonstrations. Specifically, our method is based on the variational autoencoder with a categorical latent variable. The encoder infers discrete latent factors corresponding to different behaviors from demonstrations. The decoder, as a policy, performs the behaviors accordingly. Once learned, the policy is able to reproduce a specific behavior by simply conditioning on a categorical vector. We evaluate our method on three different tasks, including a challenging task with high-dimensional visual inputs. Experimental results show that our approach is better than various baseline methods and competitive with a multi-modal policy trained by ground truth behavior labels.

Hou-Ning Hu, Qi-Zhi Cai, Dequan Wang et al.

Vehicle 3D extents and trajectories are critical cues for predicting the future location of vehicles and planning future agent ego-motion based on those predictions. In this paper, we propose a novel online framework for 3D vehicle detection and tracking from monocular videos. The framework can not only associate detections of vehicles in motion over time, but also estimate their complete 3D bounding box information from a sequence of 2D images captured on a moving platform. Our method leverages 3D box depth-ordering matching for robust instance association and utilizes 3D trajectory prediction for re-identification of occluded vehicles. We also design a motion learning module based on an LSTM for more accurate long-term motion extrapolation. Our experiments on simulation, KITTI, and Argoverse datasets show that our 3D tracking pipeline offers robust data association and tracking. On Argoverse, our image-based method is significantly better for tracking 3D vehicles within 30 meters than the LiDAR-centric baseline methods.

An-Chieh Cheng, Chieh Hubert Lin, Da-Cheng Juan et al.

Conventional Neural Architecture Search (NAS) aims at finding a single architecture that achieves the best performance, which usually optimizes task related learning objectives such as accuracy. However, a single architecture may not be representative enough for the whole dataset with high diversity and variety. Intuitively, electing domain-expert architectures that are proficient in domain-specific features can further benefit architecture related objectives such as latency. In this paper, we propose InstaNAS---an instance-aware NAS framework---that employs a controller trained to search for a "distribution of architectures" instead of a single architecture; This allows the model to use sophisticated architectures for the difficult samples, which usually comes with large architecture related cost, and shallow architectures for those easy samples. During the inference phase, the controller assigns each of the unseen input samples with a domain expert architecture that can achieve high accuracy with customized inference costs. Experiments within a search space inspired by MobileNetV2 show InstaNAS can achieve up to 48.8% latency reduction without compromising accuracy on a series of datasets against MobileNetV2.

Cheng Kuan Chen, Zhu Feng Pan, Min Sun et al.

Most of the existing works on image description focus on generating expressive descriptions. The only few works that are dedicated to generating stylish (e.g., romantic, lyric, etc.) descriptions suffer from limited style variation and content digression. To address these limitations, we propose a controllable stylish image description generation model. It can learn to generate stylish image descriptions that are more related to image content and can be trained with the arbitrary monolingual corpus without collecting new paired image and stylish descriptions. Moreover, it enables users to generate various stylish descriptions by plugging in style-specific parameters to include new styles into the existing model. We achieve this capability via a novel layer normalization layer design, which we will refer to as the Domain Layer Norm (DLN). Extensive experimental validation and user study on various stylish image description generation tasks are conducted to show the competitive advantages of the proposed model.

An-Chieh Cheng, Jin-Dong Dong, Chi-Hung Hsu et al.

Recent breakthroughs in Neural Architectural Search (NAS) have achieved state-of-the-art performance in many tasks such as image classification and language understanding. However, most existing works only optimize for model accuracy and largely ignore other important factors imposed by the underlying hardware and devices, such as latency and energy, when making inference. In this paper, we first introduce the problem of NAS and provide a survey on recent works. Then we deep dive into two recent advancements on extending NAS into multiple-objective frameworks: MONAS and DPP-Net. Both MONAS and DPP-Net are capable of optimizing accuracy and other objectives imposed by devices, searching for neural architectures that can be best deployed on a wide spectrum of devices: from embedded systems and mobile devices to workstations. Experimental results are poised to show that architectures found by MONAS and DPP-Net achieves Pareto optimality w.r.t the given objectives for various devices.

Tz-Ying Wu, Juan-Ting Lin, Tsun-Hsuang Wang et al.

Humans have the amazing ability to perform very subtle manipulation task using a closed-loop control system with imprecise mechanics (i.e., our body parts) but rich sensory information (e.g., vision, tactile, etc.). In the closed-loop system, the ability to monitor the state of the task via rich sensory information is important but often less studied. In this work, we take liquid pouring as a concrete example and aim at learning to continuously monitor whether liquid pouring is successful (e.g., no spilling) or not via rich sensory inputs. We mimic humans' rich sensories using synchronized observation from a chest-mounted camera and a wrist-mounted IMU sensor. Given many success and failure demonstrations of liquid pouring, we train a hierarchical LSTM with late fusion for monitoring. To improve the robustness of the system, we propose two auxiliary tasks during training: inferring (1) the initial state of containers and (2) forecasting the one-step future 3D trajectory of the hand with an adversarial training procedure. These tasks encourage our method to learn representation sensitive to container states and how objects are manipulated in 3D. With these novel components, our method achieves ~8% and ~11% better monitoring accuracy than the baseline method without auxiliary tasks on unseen containers and unseen users respectively.

Yu-Ting Chen, Wen-Yen Chang, Hai-Lun Lu et al.

Despite many advances in deep-learning based semantic segmentation, performance drop due to distribution mismatch is often encountered in the real world. Recently, a few domain adaptation and active learning approaches have been proposed to mitigate the performance drop. However, very little attention has been made toward leveraging information in videos which are naturally captured in most camera systems. In this work, we propose to leverage "motion prior" in videos for improving human segmentation in a weakly-supervised active learning setting. By extracting motion information using optical flow in videos, we can extract candidate foreground motion segments (referred to as motion prior) potentially corresponding to human segments. We propose to learn a memory-network-based policy model to select strong candidate segments (referred to as strong motion prior) through reinforcement learning. The selected segments have high precision and are directly used to finetune the model. In a newly collected surveillance camera dataset and a publicly available UrbanStreet dataset, our proposed method improves the performance of human segmentation across multiple scenes and modalities (i.e., RGB to Infrared (IR)). Last but not least, our method is empirically complementary to existing domain adaptation approaches such that additional performance gain is achieved by combining our weakly-supervised active learning approach with domain adaptation approaches.

Po-Yu Huang, Wan-Ting Hsu, Chun-Yueh Chiu et al.

Uncertainty estimation in deep learning becomes more important recently. A deep learning model can't be applied in real applications if we don't know whether the model is certain about the decision or not. Some literature proposes the Bayesian neural network which can estimate the uncertainty by Monte Carlo Dropout (MC dropout). However, MC dropout needs to forward the model $N$ times which results in $N$ times slower. For real-time applications such as a self-driving car system, which needs to obtain the prediction and the uncertainty as fast as possible, so that MC dropout becomes impractical. In this work, we propose the region-based temporal aggregation (RTA) method which leverages the temporal information in videos to simulate the sampling procedure. Our RTA method with Tiramisu backbone is 10x faster than the MC dropout with Tiramisu backbone ($N=5$). Furthermore, the uncertainty estimation obtained by our RTA method is comparable to MC dropout's uncertainty estimation on pixel-level and frame-level metrics.

Jin-Dong Dong, An-Chieh Cheng, Da-Cheng Juan et al.

Recent breakthroughs in Neural Architectural Search (NAS) have achieved state-of-the-art performances in applications such as image classification and language modeling. However, these techniques typically ignore device-related objectives such as inference time, memory usage, and power consumption. Optimizing neural architecture for device-related objectives is immensely crucial for deploying deep networks on portable devices with limited computing resources. We propose DPP-Net: Device-aware Progressive Search for Pareto-optimal Neural Architectures, optimizing for both device-related (e.g., inference time and memory usage) and device-agnostic (e.g., accuracy and model size) objectives. DPP-Net employs a compact search space inspired by current state-of-the-art mobile CNNs, and further improves search efficiency by adopting progressive search (Liu et al. 2017). Experimental results on CIFAR-10 are poised to demonstrate the effectiveness of Pareto-optimal networks found by DPP-Net, for three different devices: (1) a workstation with Titan X GPU, (2) NVIDIA Jetson TX1 embedded system, and (3) mobile phone with ARM Cortex-A53. Compared to CondenseNet and NASNet (Mobile), DPP-Net achieves better performances: higher accuracy and shorter inference time on various devices. Additional experimental results show that models found by DPP-Net also achieve considerably-good performance on ImageNet as well.

Hsien-Tzu Cheng, Chun-Hung Chao, Jin-Dong Dong et al.

Automatic saliency prediction in 360° videos is critical for viewpoint guidance applications (e.g., Facebook 360 Guide). We propose a spatial-temporal network which is (1) weakly-supervised trained and (2) tailor-made for 360° viewing sphere. Note that most existing methods are less scalable since they rely on annotated saliency map for training. Most importantly, they convert 360° sphere to 2D images (e.g., a single equirectangular image or multiple separate Normal Field-of-View (NFoV) images) which introduces distortion and image boundaries. In contrast, we propose a simple and effective Cube Padding (CP) technique as follows. Firstly, we render the 360° view on six faces of a cube using perspective projection. Thus, it introduces very little distortion. Then, we concatenate all six faces while utilizing the connectivity between faces on the cube for image padding (i.e., Cube Padding) in convolution, pooling, convolutional LSTM layers. In this way, CP introduces no image boundary while being applicable to almost all Convolutional Neural Network (CNN) structures. To evaluate our method, we propose Wild-360, a new 360° video saliency dataset, containing challenging videos with saliency heatmap annotations. In experiments, our method outperforms baseline methods in both speed and quality.

Wan-Ting Hsu, Chieh-Kai Lin, Ming-Ying Lee et al.

We propose a unified model combining the strength of extractive and abstractive summarization. On the one hand, a simple extractive model can obtain sentence-level attention with high ROUGE scores but less readable. On the other hand, a more complicated abstractive model can obtain word-level dynamic attention to generate a more readable paragraph. In our model, sentence-level attention is used to modulate the word-level attention such that words in less attended sentences are less likely to be generated. Moreover, a novel inconsistency loss function is introduced to penalize the inconsistency between two levels of attentions. By end-to-end training our model with the inconsistency loss and original losses of extractive and abstractive models, we achieve state-of-the-art ROUGE scores while being the most informative and readable summarization on the CNN/Daily Mail dataset in a solid human evaluation.

Tsun-Hsuan Wang, Hung-Jui Huang, Juan-Ting Lin et al.

$ $Visual place recognition is challenging, especially when only a few place exemplars are given. To mitigate the challenge, we consider place recognition method using omnidirectional cameras and propose a novel Omnidirectional Convolutional Neural Network (O-CNN) to handle severe camera pose variation. Given a visual input, the task of the O-CNN is not to retrieve the matched place exemplar, but to retrieve the closest place exemplar and estimate the relative distance between the input and the closest place. With the ability to estimate relative distance, a heuristic policy is proposed to navigate a robot to the retrieved closest place. Note that the network is designed to take advantage of the omnidirectional view by incorporating circular padding and rotation invariance. To train a powerful O-CNN, we build a virtual world for training on a large scale. We also propose a continuous lifted structured feature embedding loss to learn the concept of distance efficiently. Finally, our experimental results confirm that our method achieves state-of-the-art accuracy and speed with both the virtual world and real-world datasets.

Shih-Han Chou, Yi-Chun Chen, Kuo-Hao Zeng et al.

Narrated 360° videos are typically provided in many touring scenarios to mimic real-world experience. However, previous work has shown that smart assistance (i.e., providing visual guidance) can significantly help users to follow the Normal Field of View (NFoV) corresponding to the narrative. In this project, we aim at automatically grounding the NFoVs of a 360° video given subtitles of the narrative (referred to as "NFoV-grounding"). We propose a novel Visual Grounding Model (VGM) to implicitly and efficiently predict the NFoVs given the video content and subtitles. Specifically, at each frame, we efficiently encode the panorama into feature map of candidate NFoVs using a Convolutional Neural Network (CNN) and the subtitles to the same hidden space using an RNN with Gated Recurrent Units (GRU). Then, we apply soft-attention on candidate NFoVs to trigger sentence decoder aiming to minimize the reconstruct loss between the generated and given sentence. Finally, we obtain the NFoV as the candidate NFoV with the maximum attention without any human supervision. To train VGM more robustly, we also generate a reverse sentence conditioning on one minus the soft-attention such that the attention focuses on candidate NFoVs less relevant to the given sentence. The negative log reconstruction loss of the reverse sentence (referred to as "irrelevant loss") is jointly minimized to encourage the reverse sentence to be different from the given sentence. To evaluate our method, we collect the first narrated 360° videos dataset and achieve state-of-the-art NFoV-grounding performance.

Tz-Ying Wu, Ting-An Chien, Cheng-Sheng Chan et al.

Anticipating human intention by observing one's actions has many applications. For instance, picking up a cellphone, then a charger (actions) implies that one wants to charge the cellphone (intention). By anticipating the intention, an intelligent system can guide the user to the closest power outlet. We propose an on-wrist motion triggered sensing system for anticipating daily intentions, where the on-wrist sensors help us to persistently observe one's actions. The core of the system is a novel Recurrent Neural Network (RNN) and Policy Network (PN), where the RNN encodes visual and motion observation to anticipate intention, and the PN parsimoniously triggers the process of visual observation to reduce computation requirement. We jointly trained the whole network using policy gradient and cross-entropy loss. To evaluate, we collect the first daily "intention" dataset consisting of 2379 videos with 34 intentions and 164 unique action sequences. Our method achieves 92.68%, 90.85%, 97.56% accuracy on three users while processing only 29% of the visual observation on average.

Yen-Chen Lin, Ming-Yu Liu, Min Sun et al.

Deep reinforcement learning has shown promising results in learning control policies for complex sequential decision-making tasks. However, these neural network-based policies are known to be vulnerable to adversarial examples. This vulnerability poses a potentially serious threat to safety-critical systems such as autonomous vehicles. In this paper, we propose a defense mechanism to defend reinforcement learning agents from adversarial attacks by leveraging an action-conditioned frame prediction module. Our core idea is that the adversarial examples targeting at a neural network-based policy are not effective for the frame prediction model. By comparing the action distribution produced by a policy from processing the current observed frame to the action distribution produced by the same policy from processing the predicted frame from the action-conditioned frame prediction module, we can detect the presence of adversarial examples. Beyond detecting the presence of adversarial examples, our method allows the agent to continue performing the task using the predicted frame when the agent is under attack. We evaluate the performance of our algorithm using five games in Atari 2600. Our results demonstrate that the proposed defense mechanism achieves favorable performance against baseline algorithms in detecting adversarial examples and in earning rewards when the agents are under attack.

Kuo-Hao Zeng, Shih-Han Chou, Fu-Hsiang Chan et al.

For survival, a living agent must have the ability to assess risk (1) by temporally anticipating accidents before they occur, and (2) by spatially localizing risky regions in the environment to move away from threats. In this paper, we take an agent-centric approach to study the accident anticipation and risky region localization tasks. We propose a novel soft-attention Recurrent Neural Network (RNN) which explicitly models both spatial and appearance-wise non-linear interaction between the agent triggering the event and another agent or static-region involved. In order to test our proposed method, we introduce the Epic Fail (EF) dataset consisting of 3000 viral videos capturing various accidents. In the experiments, we evaluate the risk assessment accuracy both in the temporal domain (accident anticipation) and spatial domain (risky region localization) on our EF dataset and the Street Accident (SA) dataset. Our method consistently outperforms other baselines on both datasets.

Hou-Ning Hu, Yen-Chen Lin, Ming-Yu Liu et al.

Watching a 360° sports video requires a viewer to continuously select a viewing angle, either through a sequence of mouse clicks or head movements. To relieve the viewer from this "360 piloting" task, we propose "deep 360 pilot" -- a deep learning-based agent for piloting through 360° sports videos automatically. At each frame, the agent observes a panoramic image and has the knowledge of previously selected viewing angles. The task of the agent is to shift the current viewing angle (i.e. action) to the next preferred one (i.e., goal). We propose to directly learn an online policy of the agent from data. We use the policy gradient technique to jointly train our pipeline: by minimizing (1) a regression loss measuring the distance between the selected and ground truth viewing angles, (2) a smoothness loss encouraging smooth transition in viewing angle, and (3) maximizing an expected reward of focusing on a foreground object. To evaluate our method, we build a new 360-Sports video dataset consisting of five sports domains. We train domain-specific agents and achieve the best performance on viewing angle selection accuracy and transition smoothness compared to [51] and other baselines.

Tseng-Hung Chen, Yuan-Hong Liao, Ching-Yao Chuang et al.

Impressive image captioning results are achieved in domains with plenty of training image and sentence pairs (e.g., MSCOCO). However, transferring to a target domain with significant domain shifts but no paired training data (referred to as cross-domain image captioning) remains largely unexplored. We propose a novel adversarial training procedure to leverage unpaired data in the target domain. Two critic networks are introduced to guide the captioner, namely domain critic and multi-modal critic. The domain critic assesses whether the generated sentences are indistinguishable from sentences in the target domain. The multi-modal critic assesses whether an image and its generated sentence are a valid pair. During training, the critics and captioner act as adversaries -- captioner aims to generate indistinguishable sentences, whereas critics aim at distinguishing them. The assessment improves the captioner through policy gradient updates. During inference, we further propose a novel critic-based planning method to select high-quality sentences without additional supervision (e.g., tags). To evaluate, we use MSCOCO as the source domain and four other datasets (CUB-200-2011, Oxford-102, TGIF, and Flickr30k) as the target domains. Our method consistently performs well on all datasets. In particular, on CUB-200-2011, we achieve 21.8% CIDEr-D improvement after adaptation. Utilizing critics during inference further gives another 4.5% boost.

Yi-Ling Chen, Jan Klopp, Min Sun et al.

Photo composition is an important factor affecting the aesthetics in photography. However, it is a highly challenging task to model the aesthetic properties of good compositions due to the lack of globally applicable rules to the wide variety of photographic styles. Inspired by the thinking process of photo taking, we formulate the photo composition problem as a view finding process which successively examines pairs of views and determines their aesthetic preferences. We further exploit the rich professional photographs on the web to mine unlimited high-quality ranking samples and demonstrate that an aesthetics-aware deep ranking network can be trained without explicitly modeling any photographic rules. The resulting model is simple and effective in terms of its architectural design and data sampling method. It is also generic since it naturally learns any photographic rules implicitly encoded in professional photographs. The experiments show that the proposed view finding network achieves state-of-the-art performance with sliding window search strategy on two image cropping datasets.

Kuo-Hao Zeng, Tseng-Hung Chen, Ching-Yao Chuang et al.

We propose a scalable approach to learn video-based question answering (QA): answer a "free-form natural language question" about a video content. Our approach automatically harvests a large number of videos and descriptions freely available online. Then, a large number of candidate QA pairs are automatically generated from descriptions rather than manually annotated. Next, we use these candidate QA pairs to train a number of video-based QA methods extended fromMN (Sukhbaatar et al. 2015), VQA (Antol et al. 2015), SA (Yao et al. 2015), SS (Venugopalan et al. 2015). In order to handle non-perfect candidate QA pairs, we propose a self-paced learning procedure to iteratively identify them and mitigate their effects in training. Finally, we evaluate performance on manually generated video-based QA pairs. The results show that our self-paced learning procedure is effective, and the extended SS model outperforms various baselines.

Kuo-Hao Zeng, Tseng-Hung Chen, Juan Carlos Niebles et al.

A great video title describes the most salient event compactly and captures the viewer's attention. In contrast, video captioning tends to generate sentences that describe the video as a whole. Although generating a video title automatically is a very useful task, it is much less addressed than video captioning. We address video title generation for the first time by proposing two methods that extend state-of-the-art video captioners to this new task. First, we make video captioners highlight sensitive by priming them with a highlight detector. Our framework allows for jointly training a model for title generation and video highlight localization. Second, we induce high sentence diversity in video captioners, so that the generated titles are also diverse and catchy. This means that a large number of sentences might be required to learn the sentence structure of titles. Hence, we propose a novel sentence augmentation method to train a captioner with additional sentence-only examples that come without corresponding videos. We collected a large-scale Video Titles in the Wild (VTW) dataset of 18100 automatically crawled user-generated videos and titles. On VTW, our methods consistently improve title prediction accuracy, and achieve the best performance in both automatic and human evaluation. Finally, our sentence augmentation method also outperforms the baselines on the M-VAD dataset.

Cheng-Sheng Chan, Shou-Zhong Chen, Pei-Xuan Xie et al.

We revisit the study of a wrist-mounted camera system (referred to as HandCam) for recognizing activities of hands. HandCam has two unique properties as compared to egocentric systems (referred to as HeadCam): (1) it avoids the need to detect hands; (2) it more consistently observes the activities of hands. By taking advantage of these properties, we propose a deep-learning-based method to recognize hand states (free v.s. active hands, hand gestures, object categories), and discover object categories. Moreover, we propose a novel two-streams deep network to further take advantage of both HandCam and HeadCam. We have collected a new synchronized HandCam and HeadCam dataset with 20 videos captured in three scenes for hand states recognition. Experiments show that our HandCam system consistently outperforms a deep-learning-based HeadCam method (with estimated manipulation regions) and a dense-trajectory-based HeadCam method in all tasks. We also show that HandCam videos captured by different users can be easily aligned to improve free v.s. active recognition accuracy (3.3% improvement) in across-scenes use case. Moreover, we observe that finetuning Convolutional Neural Network consistently improves accuracy. Finally, our novel two-streams deep network combining HandCam and HeadCam features achieves the best performance in four out of five tasks. With more data, we believe a joint HandCam and HeadCam system can robustly log hand states in daily life.