RefineNet: Multi-Path Refinement Networks for Dense Prediction.

Recently, very deep convolutional neural networks (CNNs) have shown outstanding performance in object recognition and have also been the first choice for dense classification problems such as semantic segmentation. However, repeated subsampling operations like pooling or convolution striding in deep CNNs lead to a significant decrease in the initial image resolution. Here, we present RefineNet, a generic multi-path refinement network that explicitly exploits all the information available along the down-sampling process to enable high-resolution prediction using long-range residual connections. In this way, the deeper layers that capture high-level semantic features can be directly refined using fine-grained features from earlier convolutions. The individual components of RefineNet employ residual connections following the identity mapping mindset, which allows for effective end-to-end training. Further, we introduce chained residual pooling, which captures rich background context in an efficient manner. We carry out comprehensive experiments and set new state-of-the-art results on seven public datasets. In particular, we achieve an intersection-over-union score of 83.4 on the challenging PASCAL VOC 2012 dataset, which is the best reported result to date.

Convolutional Neural Networks (CNNs) have achieved significant success in image classification and object detection. CNN models generally consist of a single-stream and process single image data at once. In addition, multi-stream (or multi-modal) models have recently begun to be proposed that allow the processing of more than one input at the same time. The data can be an image, video, voice, or any other sensor data. Multi-modality may help us extract some hidden features of the same object. Furthermore, several new studies examine sharing feature maps between different streams of the same CNN. However, systematic studies that can adequately demonstrate the contribution of multi-modality and feature map sharing features to performance have not yet been conducted. Processing power and lack of available datasets are among the important factors that negatively affect progress. In this study, the contributions of multi-modality and feature map sharing (FMS) to increase the performance in object recognition are examined in detail. For this purpose, a new dataset and a new multi-modal multi-feature map sharing CNN model, which we call FMSNet, are developed. The proposed model achieved a 3.06% higher accuracy rate than its non-FMS counterpart, DenseNet-201, exceeding most of the state-of-the-art single-stream CNN models.

Some exiting works have explored the promotion between depth estimation and semantic segmentation. These works are usually based on convolutional neural networks, which extract compact features and map the relationship between input and output according to specific tasks. In this paper, we introduce a novel adversarial training strategy, that is, generator produces the semantic segmentation map and depth map, and then the discriminator can judge the authenticity of the synthesized color image, thereby supervising the output of the network during back propagation. We use the adversarial loss combined with the reconstruction loss function to supervise the model, and find that the adversarial loss function which is seen as a global supervision can further optimize the output. We use 200 color images from Kitti dataset with semantic segmentation ground truth as the training set, and train the network in an end-to-end manner. The experimental results show that the adversarial training method is well applied to the multi-task training combining semantic segmentation and depth estimation, and can further improve the quantitative performance of depth estimation.

Point cloud data provide three-dimensional (3D) information about objects in the real world, containing rich semantic features. Therefore, the task of semantic segmentation of point clouds has been widely applied in fields such as robotics and autonomous driving. Although existing research has made unprecedented progress, achieving real-time semantic segmentation of point clouds on airborne devices still faces challenges due to excessive computational and memory requirements. To address this issue, we propose a novel sequence convolution semantic segmentation architecture that integrates Convolutional Neural Networks (CNN) with a sequence-to-sequence (seq2seq) structure, termed SeqConv-Net. This architecture views point cloud semantic segmentation as a sequence generation task. Based on our unique perspective of spatially ordered sequences, we use Recurrent Neural Networks (RNN) to encode elevation information, then input the structured hidden states into a CNN for planar feature extraction. The results are combined with the RNN’s encoded outputs via residual connections and are fed into a decoder for sequence prediction in a seq2seq manner. Experiments show that the SeqConv-Net architecture achieves 75.5% mean Intersection Over Union (mIOU) accuracy on the DALES dataset, with the total processing speed from data preprocessing to prediction being several to tens of times faster than existing methods. Additionally, SeqConv-Net can balance accuracy and speed by adjusting the hyperparameters and using different RNNs and CNNs, providing a new solution for real-time point cloud semantic segmentation in airborne environments.

Accurate semantic segmentation of 3D point clouds is a long-standing problem in remote sensing and computer vision. Due to the unstructured nature of point clouds, designing deep neural architectures for point cloud semantic segmentation is often not straightforward. In this work, we circumvent this problem by devising a technique to exploit structured neural architectures for unstructured data. In particular, we employ the popular convolutional neural network (CNN) architectures to perform semantic segmentation of LiDAR data. We propose a projection-based scheme that performs an angle-wise slicing of large 3D point clouds and transforms those slices into 2D grids. Accounting for intensity and reflectivity of the LiDAR input, the 2D grid allows us to construct a pseudo image for the point cloud slice. We enhance this image with low-level image processing techniques of normalization, histogram equalization, and decorrelation stretch to suit our ultimate object of semantic segmentation. A large number of images thus generated are used to induce an encoder-decoder CNN model that learns to compute a segmented 2D projection of the scene, which we finally back project to the 3D point cloud. In addition to a novel method, this article also makes a second major contribution of introducing the enhanced version of our large-scale public PC-Urban outdoor dataset which is captured in a civic setup with an Ouster LiDAR sensor. The updated dataset (PC-Urban_V2) provides nearly 8 billion points including over 100 million points labeled for 25 classes of interest. We provide a thorough evaluation of our technique on PC-Urban_V2 and three other public datasets.

Recent experiments in computer vision demonstrate texture bias as the primary reason for supreme results in models employing Convolutional Neural Networks (CNNs), conflicting with early works claiming that these networks identify objects using shape. It is believed that the cost function forces the CNN to take a greedy approach and develop a proclivity for local information like texture to increase accuracy, thus failing to explore any global statistics. We propose CognitiveCNN, a new intuitive architecture, inspired from feature integration theory in psychology to utilise human-interpretable feature like shape, texture, edges etc. to reconstruct, and classify the image. We define novel metrics to quantify the "relevance" of "abstract information" present in these modalities using attention maps. We further introduce a regularisation method which ensures that each modality like shape, texture etc. gets proportionate influence in a given task, as it does for reconstruction; and perform experiments to show the resulting boost in accuracy and robustness, be-sides imparting explainability to these CNNs for achieving superior performance in object recognition.

In this study, we present a method for all-around depth estimation from multiple omnidirectional images for indoor environments. In particular, we focus on plane-sweeping stereo as the method for depth estimation from the images. We propose a new icosahedron-based representation and ConvNets for omnidirectional images, which we name "CrownConv" because the representation resembles a crown made of origami. CrownConv can be applied to both fisheye images and equirectangular images to extract features. Furthermore, we propose icosahedron-based spherical sweeping for generating the cost volume on an icosahedron from the extracted features. The cost volume is regularized using the three-dimensional CrownConv, and the final depth is obtained by depth regression from the cost volume. Our proposed method is robust to camera alignments by using the extrinsic camera parameters; therefore, it can achieve precise depth estimation even when the camera alignment differs from that in the training dataset. We evaluate the proposed model on synthetic datasets and demonstrate its effectiveness. As our proposed method is computationally efficient, the depth is estimated from four fisheye images in less than a second using a laptop with a GPU. Therefore, it is suitable for real-world robotics applications. Our source code is available at https://github.com/matsuren/crownconv360depth.

In the contemporary landscape of accelerating artificial intelligence (AI) development, multi‐dimensional information recognition has emerged as a critical enabler for enhancing both data computational efficiency and decision‐making precision. However, traditional multi‐dimensional recognition architectures exhibit a fundamental reliance on extensive hardware arrays and complex circuit topologies, posing significant challenges to hardware integration and system‐level miniaturization for AI‐based recognition systems. Here, for the first time, we propose an in situ 4D neuromorphic transistor (I‐FNT) and design a 4D spatiotemporal recognition system based on I‐FNT. Through dynamic encoding of the input port voltages of I‐FNT, programmable switching among three recognition modes (grayscale, depth, and time) is achieved, enabling cross‐dimensional information perception. Compared to existing multi‐dimensional information recognition systems, our 4D spatiotemporal recognition system significantly simplifies hardware while achieving 100% device integration gain. The I‐FNT‐integrated convolutional neural network (CNN) harnesses spatial (depth) information to achieve breakthrough performance in object recognition: 122% higher training efficiency and 345% faster training speed relative to conventional architectures, while attaining 94% accuracy. The system simultaneously facilitates object motion trajectory recognition, demonstrating comprehensive spatiotemporal processing capabilities. Therefore, I‐FNT provides an efficient and accurate novel solution for multi‐dimensional information recognition, representing a significant breakthrough for intelligent sensing and AI‐based recognition systems.

Single image depth estimation works fail to separate foreground elements because they can easily be confounded with the background. To alleviate this problem, we propose the use of a semantic segmentation procedure that adds information to a depth estimator, in this case, a 3D Convolutional Neural Network (CNN)—segmentation is coded as one-hot planes representing categories of objects. We explore 2D and 3D models. Particularly, we propose a hybrid 2D–3D CNN architecture capable of obtaining semantic segmentation and depth estimation at the same time. We tested our procedure on the SYNTHIA-AL dataset and obtained , which is an improvement of 0.14 points (compared with the state of the art of ) by using manual segmentation, and using automatic semantic segmentation, proving that depth estimation is improved when the shape and position of objects in a scene are known.

Deep convolutional neural networks (CNNs) have been widely and successfully applied in many computer vision tasks, such as classification, detection, semantic segmentation, and so on. As for image retrieval, while off-the-shelf CNN features from models trained for classification task are demonstrated promising, it remains a challenge to learn specific features oriented for instance retrieval. Witnessing the great success of low-level SIFT feature in image retrieval and its complementary nature to the semantic-aware CNN feature, in this paper, we propose to embed the SIFT feature into the CNN feature with a Siamese structure in a learning-based paradigm. The learning objective consists of two kinds of loss, i.e., similarity loss and fidelity loss. The first loss embeds the image-level nearest neighborhood structure with the SIFT feature into CNN feature learning, while the second loss imposes that the CNN feature with the updated CNN model preserves the fidelity of that from the original CNN model solely trained for classification. After the learning, the generated CNN feature inherits the property of the SIFT feature, which is well oriented for image retrieval. We evaluate our approach on the public data sets, and comprehensive experiments demonstrate the effectiveness of the proposed method.

Object classification has been one of the main tasks in computer vision. With the fast development of deep learning, its performance in image classification and object recognition has presented dramatic improvements. In this paper, we propose a new deep convolutional neural network (CNN) architecture for robust object classification. The proposed model is fused with three traditional CNN approaches, Densenet201, Resnet50, and our proposed residual CNN. The fused network architecture allows parallel processing of the multiple networks for keeping the system sped up. A single shot deep convolution network is trained as an object detector to generate all possible candidates of different object classes. The output of each neural network is representing a single vote that is used in the classification process. 3-to-1 voting criteria are applied in the final classification decision between the candidate object classes. Several experiments were conducted to evaluate the performance of the proposed network. The experimental results show that the proposed approach has better performance than the networks used in the fusion process when they act individually. It also has lower miss rates when compared to several states of art methodologies.

Image semantic segmentation contains two sub-tasks, segmenting and labeling. However, the recent fully convolutional network (FCN) based methods often ignore the first sub-task and consider it as a direct labeling one. Even though these methods have achieved competitive performances, they obtained spatially fragmented and disconnected outputs. The reason is that, pixel-level relationships inside the deepest layers become inconsistent since traditional FCNs do not have any explicit pixel grouping mechanism. To address this problem, a multi-step supervised learning method, which contains image-level supervised learning step and pixel-level supervised learning step, is proposed. Specifically, as for the visualized result of image semantic segmentation, it is actually an image-to-image transformation problem, from RGB domain to category label domain. The recent conditional generative adversarial network (cGAN) has achieved significant performance for image-to-image generation task, and the generated image remains good regional connectivity. Therefore, a cGAN supervised by RGB-category label map is used to obtain a coarse segmentation mask, which avoids generating disconnected segmentation results to a certain extent. Furthermore, an interaction information (II) loss term is proposed for cGAN to remain the spatial structure of the segmentation mask. Additionally, dilated convolutional networks (DCNs) have achieved significant performance in object detection field, especially for small objects because of its special receptive field settings. Specific to image semantic segmentation, if each pixel is seen as an object, this task can be transformed to object detection. In this case, combined with the segmentation mask from cGAN, a DCN supervised by the pixel-level label is used to finalize the category recognition of each pixel in the image. The proposed method achieves satisfactory performances on three public and challenging datasets for image semantic segmentation.

Pixel-level labelling tasks, such as semantic segmentation, play a central role in image understanding. Recent approaches have attempted to harness the capabilities of deep learning techniques for image recognition to tackle pixel-level labelling tasks. One central issue in this methodology is the limited capacity of deep learning techniques to delineate visual objects. To solve this problem, we introduce a new form of convolutional neural network that combines the strengths of Convolutional Neural Networks (CNNs) and Conditional Random Fields (CRFs)-based probabilistic graphical modelling. To this end, we formulate mean-field approximate inference for the Conditional Random Fields with Gaussian pairwise potentials as Recurrent Neural Networks. This network, called CRF-RNN, is then plugged in as a part of a CNN to obtain a deep network that has desirable properties of both CNNs and CRFs. Importantly, our system fully integrates CRF modelling with CNNs, making it possible to train the whole deep network end-to-end with the usual back-propagation algorithm, avoiding offline post-processing methods for object delineation. We apply the proposed method to the problem of semantic image segmentation, obtaining top results on the challenging Pascal VOC 2012 segmentation benchmark.

Colorectal cancer typically affects the gastrointestinal tract within the human body. Colonoscopy is one of the most accurate methods of detecting cancer. The current system facilitates the identification of cancer by computer-assisted diagnosis (CADx) systems with a limited number of deep learning methods. It does not imply the depiction of mixed datasets for the functioning of the system. The proposed system, called ColoRectalCADx, is supported by deep learning (DL) models suitable for cancer research. The CADx system comprises five stages: convolutional neural networks (CNN), support vector machine (SVM), long short-term memory (LSTM), visual explanation such as gradient-weighted class activation mapping (Grad-CAM), and semantic segmentation phases. Here, the key components of the CADx system are equipped with 9 individual and 12 integrated CNNs, implying that the system consists mainly of investigational experiments with a total of 21 CNNs. In the subsequent phase, the CADx has a combination of CNNs of concatenated transfer learning functions associated with the machine SVM classification. Additional classification is applied to ensure effective transfer of results from CNN to LSTM. The system is mainly made up of a combination of CVC Clinic DB, Kvasir2, and Hyper Kvasir input as a mixed dataset. After CNN and LSTM, in advanced stage, malignancies are detected by using a better polyp recognition technique with Grad-CAM and semantic segmentation using U-Net. CADx results have been stored on Google Cloud for record retention. In these experiments, among all the CNNs, the individual CNN DenseNet-201 (87.1% training and 84.7% testing accuracies) and the integrated CNN ADaDR-22 (84.61% training and 82.17% testing accuracies) were the most efficient for cancer detection with the CNN+LSTM model. ColoRectalCADx accurately identifies cancer through individual CNN DesnseNet-201 and integrated CNN ADaDR-22. In Grad-CAM's visual explanations, CNN DenseNet-201 displays precise visualization of polyps, and CNN U-Net provides precise malignant polyps.

Morphological attributes of human blastocyst components and their characteristics are highly correlated with the success rate of in vitro fertilization (IVF). Blastocyst component analysis aims to choose the most viable embryos to improve the success rate of IVF. The embryologist evaluates blastocyst viability by manual microscopic assessment of its components, such as zona pellucida (ZP), trophectoderm (TE), blastocoel (BL), and inner cell mass (ICM). With the success of deep learning in the medical diagnosis domain, semantic segmentation has the potential to detect crucial components of human blastocysts for computerized analysis. In this study, a sprint semantic segmentation network (SSS-Net) is proposed to accurately detect blastocyst components for embryological analysis. The proposed method is based on a fully convolutional semantic segmentation scheme that provides the pixel-wise classification of important blastocyst components that help to automatically check the morphologies of these elements. The proposed SSS-Net uses the sprint convolutional block (SCB), which uses asymmetric kernel convolutions in combination with depth-wise separable convolutions to reduce the overall cost of the network. SSS-Net is a shallow architecture with dense feature aggregation, which helps in better segmentation. The proposed SSS-Net consumes a smaller number of trainable parameters (4.04 million) compared to state-of-the-art methods. The SSS-Net was evaluated using a publicly available human blastocyst image dataset for component segmentation. The experimental results confirm that our proposal provides promising segmentation performance with a Jaccard Index of 82.88%, 77.40%, 88.39%, 84.94%, and 96.03% for ZP, TE, BL, ICM, and background, with residual connectivity, respectively. It is also provides a Jaccard Index of 84.51%, 78.15%, 88.68%, 84.50%, and 95.82% for ZP, TE, BL, ICM, and background, with dense connectivity, respectively. The proposed SSS-Net is providing a mean Jaccard Index (Mean JI) of 85.93% and 86.34% with residual and dense connectivity, respectively; this shows effective segmentation of blastocyst components for embryological analysis.