Nilanjan Ray

Seyed Mojtaba Marvasti-Zadeh, Devin Goodsman, Nilanjan Ray et al.

This paper provides a comprehensive review of past and current advances in the early detection of bark beetle-induced tree mortality from three primary perspectives: bark beetle & host interactions, RS, and ML/DL. In contrast to prior efforts, this review encompasses all RS systems and emphasizes ML/DL methods to investigate their strengths and weaknesses. We parse existing literature based on multi- or hyper-spectral analyses and distill their knowledge based on: bark beetle species & attack phases with a primary emphasis on early stages of attacks, host trees, study regions, RS platforms & sensors, spectral/spatial/temporal resolutions, spectral signatures, spectral vegetation indices (SVIs), ML approaches, learning schemes, task categories, models, algorithms, classes/clusters, features, and DL networks & architectures. Although DL-based methods and the random forest (RF) algorithm showed promising results, highlighting their potential to detect subtle changes across visible, thermal, and short-wave infrared (SWIR) spectral regions, they still have limited effectiveness and high uncertainties. To inspire novel solutions to these shortcomings, we delve into the principal challenges & opportunities from different perspectives, enabling a deeper understanding of the current state of research and guiding future research directions.

Rudraksh Kapil, Seyed Mojtaba Marvasti-Zadeh, Nadir Erbilgin et al.

Accurate detection of individual tree crowns from remote sensing data poses a significant challenge due to the dense nature of forest canopy and the presence of diverse environmental variations, e.g., overlapping canopies, occlusions, and varying lighting conditions. Additionally, the lack of data for training robust models adds another limitation in effectively studying complex forest conditions. This paper presents a novel method for detecting shadowed tree crowns and provides a challenging dataset comprising roughly 50k paired RGB-thermal images to facilitate future research for illumination-invariant detection. The proposed method (ShadowSense) is entirely self-supervised, leveraging domain adversarial training without source domain annotations for feature extraction and foreground feature alignment for feature pyramid networks to adapt domain-invariant representations by focusing on visible foreground regions, respectively. It then fuses complementary information of both modalities to effectively improve upon the predictions of an RGB-trained detector and boost the overall accuracy. Extensive experiments demonstrate the superiority of the proposed method over both the baseline RGB-trained detector and state-of-the-art techniques that rely on unsupervised domain adaptation or early image fusion. Our code and data are available: https://github.com/rudrakshkapil/ShadowSense

Rudraksh Kapil, Seyed Mojtaba Marvasti-Zadeh, Devin Goodsman et al.

Bark beetle outbreaks can dramatically impact forest ecosystems and services around the world. For the development of effective forest policies and management plans, the early detection of infested trees is essential. Despite the visual symptoms of bark beetle infestation, this task remains challenging, considering overlapping tree crowns and non-homogeneity in crown foliage discolouration. In this work, a deep learning based method is proposed to effectively classify different stages of bark beetle attacks at the individual tree level. The proposed method uses RetinaNet architecture (exploiting a robust feature extraction backbone pre-trained for tree crown detection) to train a shallow subnetwork for classifying the different attack stages of images captured by unmanned aerial vehicles (UAVs). Moreover, various data augmentation strategies are examined to address the class imbalance problem, and consequently, the affine transformation is selected to be the most effective one for this purpose. Experimental evaluations demonstrate the effectiveness of the proposed method by achieving an average accuracy of 98.95%, considerably outperforming the baseline method by approximately 10%.

Amir Akbarnejad, Nilanjan Ray, Penny J. Barnes et al.

Despite the advances in machine learning and digital pathology, it is not yet clear if machine learning methods can accurately predict molecular information merely from histomorphology. In a quest to answer this question, we built a large-scale dataset (185538 images) with reliable measurements for Ki67, ER, PR, and HER2 statuses. The dataset is composed of mirrored images of H\&E and corresponding images of immunohistochemistry (IHC) assays (Ki67, ER, PR, and HER2. These images are mirrored through registration. To increase reliability, individual pairs were inspected and discarded if artifacts were present (tissue folding, bubbles, etc). Measurements for Ki67, ER and PR were determined by calculating H-Score from image analysis. HER2 measurement is based on binary classification: 0 and 1+ (IHC scores representing a negative subset) vs 3+ (IHC score positive subset). Cases with IHC equivocal score (2+) were excluded. We show that a standard ViT-based pipeline can achieve prediction performances around 90% in terms of Area Under the Curve (AUC) when trained with a proper labeling protocol. Finally, we shed light on the ability of the trained classifiers to localize relevant regions, which encourages future work to improve the localizations. Our proposed dataset is publicly available: https://ihc4bc.github.io/

Seyed Mojtaba Marvasti-Zadeh, Devin Goodsman, Nilanjan Ray et al.

Visual explanation of ``black-box'' models allows researchers in explainable artificial intelligence (XAI) to interpret the model's decisions in a human-understandable manner. In this paper, we propose interpretable class activation mapping for tree crown detection (Crown-CAM) that overcomes inaccurate localization & computational complexity of previous methods while generating reliable visual explanations for the challenging and dynamic problem of tree crown detection in aerial images. It consists of an unsupervised selection of activation maps, computation of local score maps, and non-contextual background suppression to efficiently provide fine-grain localization of tree crowns in scenarios with dense forest trees or scenes without tree crowns. Additionally, two Intersection over Union (IoU)-based metrics are introduced to effectively quantify both the accuracy and inaccuracy of generated explanations with respect to regions with or even without tree crowns in the image. Empirical evaluations demonstrate that the proposed Crown-CAM outperforms the Score-CAM, Augmented Score-CAM, and Eigen-CAM methods by an average IoU margin of 8.7, 5.3, and 21.7 (and 3.3, 9.8, and 16.5) respectively in improving the accuracy (and decreasing inaccuracy) of visual explanations on the challenging NEON tree crown dataset.

Preetam Anbukarasu, Shailesh Nanisetty, Ganesh Tata et al.

The focus of this paper is a proof of concept, machine learning (ML) pipeline that extracts heart rate from pressure sensor data acquired on low-power edge devices. The ML pipeline consists an upsampler neural network, a signal quality classifier, and a 1D-convolutional neural network optimized for efficient and accurate heart rate estimation. The models were designed so the pipeline was less than 40 kB. Further, a hybrid pipeline consisting of the upsampler and classifier, followed by a peak detection algorithm was developed. The pipelines were deployed on ESP32 edge device and benchmarked against signal processing to determine the energy usage, and inference times. The results indicate that the proposed ML and hybrid pipeline reduces energy and time per inference by 82% and 28% compared to traditional algorithms. The main trade-off for ML pipeline was accuracy, with a mean absolute error (MAE) of 3.28, compared to 2.39 and 1.17 for the hybrid and signal processing pipelines. The ML models thus show promise for deployment in energy and computationally constrained devices. Further, the lower sampling rate and computational requirements for the ML pipeline could enable custom hardware solutions to reduce the cost and energy needs of wearable devices.

Seyed Mojtaba Marvasti-Zadeh, Nilanjan Ray, Nadir Erbilgin

Semi-supervised object detection (SSOD) aims to improve the performance and generalization of existing object detectors by utilizing limited labeled data and extensive unlabeled data. Despite many advances, recent SSOD methods are still challenged by inadequate model refinement using the classical exponential moving average (EMA) strategy, the consensus of Teacher-Student models in the latter stages of training (i.e., losing their distinctiveness), and noisy/misleading pseudo-labels. This paper proposes a novel training-based model refinement (TMR) stage and a simple yet effective representation disagreement (RD) strategy to address the limitations of classical EMA and the consensus problem. The TMR stage of Teacher-Student models optimizes the lightweight scaling operation to refine the model's weights and prevent overfitting or forgetting learned patterns from unlabeled data. Meanwhile, the RD strategy helps keep these models diverged to encourage the student model to explore additional patterns in unlabeled data. Our approach can be integrated into established SSOD methods and is empirically validated using two baseline methods, with and without cascade regression, to generate more reliable pseudo-labels. Extensive experiments demonstrate the superior performance of our approach over state-of-the-art SSOD methods. Specifically, the proposed approach outperforms the baseline Unbiased-Teacher-v2 (& Unbiased-Teacher-v1) method by an average mAP margin of 2.23, 2.1, and 3.36 (& 2.07, 1.9, and 3.27) on COCO-standard, COCO-additional, and Pascal VOC datasets, respectively.

Ameneh Sheikhjafari, Deepa Krishnaswamy, Michelle Noga et al.

This study proposes an end-to-end unsupervised diffeomorphic deformable registration framework based on moving mesh parameterization. Using this parameterization, a deformation field can be modeled with its transformation Jacobian determinant and curl of end velocity field. The new model of the deformation field has three important advantages; firstly, it relaxes the need for an explicit regularization term and the corresponding weight in the cost function. The smoothness is implicitly embedded in the solution which results in a physically plausible deformation field. Secondly, it guarantees diffeomorphism through explicit constraints applied to the transformation Jacobian determinant to keep it positive. Finally, it is suitable for cardiac data processing, since the nature of this parameterization is to define the deformation field in terms of the radial and rotational components. The effectiveness of the algorithm is investigated by evaluating the proposed method on three different data sets including 2D and 3D cardiac MRI scans. The results demonstrate that the proposed framework outperforms existing learning-based and non-learning-based methods while generating diffeomorphic transformations.

Fateme Bahri, Nilanjan Ray

Background subtraction is a fundamental task in computer vision with numerous real-world applications, ranging from object tracking to video surveillance. Dynamic backgrounds poses a significant challenge here. Supervised deep learning-based techniques are currently considered state-of-the-art for this task. However, these methods require pixel-wise ground-truth labels, which can be time-consuming and expensive. In this work, we propose a weakly supervised framework that can perform background subtraction without requiring per-pixel ground-truth labels. Our framework is trained on a moving object-free sequence of images and comprises two networks. The first network is an autoencoder that generates background images and prepares dynamic background images for training the second network. The dynamic background images are obtained by thresholding the background-subtracted images. The second network is a U-Net that uses the same object-free video for training and the dynamic background images as pixel-wise ground-truth labels. During the test phase, the input images are processed by the autoencoder and U-Net, which generate background and dynamic background images, respectively. The dynamic background image helps remove dynamic motion from the background-subtracted image, enabling us to obtain a foreground image that is free of dynamic artifacts. To demonstrate the effectiveness of our method, we conducted experiments on selected categories of the CDnet 2014 dataset and the I2R dataset. Our method outperformed all top-ranked unsupervised methods. We also achieved better results than one of the two existing weakly supervised methods, and our performance was similar to the other. Our proposed method is online, real-time, efficient, and requires minimal frame-level annotation, making it suitable for a wide range of real-world applications.

Jakaria Rabbi, Nilanjan Ray, Dana Cobzas

Disentangling pathological changes from physiological aging in 3D medical shapes is crucial for developing interpretable biomarkers and patient stratification. However, this separation is challenging when diagnosis labels are limited or unavailable, since disease and aging often produce overlapping effects on shape changes, obscuring clinically relevant shape patterns. To address this challenge, we propose a two-stage framework combining unsupervised disease discovery with self-supervised disentanglement of implicit shape representations. In the first stage, we train an implicit neural model with signed distance functions to learn stable shape embeddings. We then apply clustering on the shape latent space, which yields pseudo disease labels without using ground-truth diagnosis during discovery. In the second stage, we disentangle factors in a compact variational space using pseudo disease labels discovered in the first stage and the ground truth age labels available for all subjects. We enforce separation and controllability with a multi-objective disentanglement loss combining covariance and a supervised contrastive loss. On ADNI hippocampus and OAI distal femur shapes, we achieve near-supervised performance, improving disentanglement and reconstruction over state-of-the-art unsupervised baselines, while enabling high-fidelity reconstruction, controllable synthesis, and factor-based explainability. Code and checkpoints are available at https://github.com/anonymous-submission01/medical-shape-disentanglement

Seyed Mojtaba Marvasti-Zadeh, Mohammad N. S. Jahromi, Javad Khaghani et al.

In nature, the collective behavior of animals, such as flying birds is dominated by the interactions between individuals of the same species. However, the study of such behavior among the bird species is a complex process that humans cannot perform using conventional visual observational techniques such as focal sampling in nature. For social animals such as birds, the mechanism of group formation can help ecologists understand the relationship between social cues and their visual characteristics over time (e.g., pose and shape). But, recovering the varying pose and shapes of flying birds is a highly challenging problem. A widely-adopted solution to tackle this bottleneck is to extract the pose and shape information from 2D image to 3D correspondence. Recent advances in 3D vision have led to a number of impressive works on the 3D shape and pose estimation, each with different pros and cons. To the best of our knowledge, this work is the first attempt to provide an overview of recent advances in 3D bird reconstruction based on monocular vision, give both computer vision and biology researchers an overview of existing approaches, and compare their characteristics.

Ganesh Tata, Katyani Singh, Eric Van Oeveren et al.

Recent work has shown that by approximating the behaviour of a non-differentiable black-box function using a neural network, the black-box can be integrated into a differentiable training pipeline for end-to-end training. This methodology is termed "differentiable bypass,'' and a successful application of this method involves training a document preprocessor to improve the performance of a black-box OCR engine. However, a good approximation of an OCR engine requires querying it for all samples throughout the training process, which can be computationally and financially expensive. Several zeroth-order optimization (ZO) algorithms have been proposed in black-box attack literature to find adversarial examples for a black-box model by computing its gradient in a query-efficient manner. However, the query complexity and convergence rate of such algorithms makes them infeasible for our problem. In this work, we propose two sample selection algorithms to train an OCR preprocessor with less than 10% of the original system's OCR engine queries, resulting in more than 60% reduction of the total training time without significant loss of accuracy. We also show an improvement of 4% in the word-level accuracy of a commercial OCR engine with only 2.5% of the total queries and a 32x reduction in monetary cost. Further, we propose a simple ranking technique to prune 30% of the document images from the training dataset without affecting the system's performance.

Arnisha Khondaker, Nilanjan Ray

Tuning parameters of a non-differentiable or black-box compute is challenging. Existing methods rely mostly on random sampling or grid sampling from the parameter space. Further, with all the current methods, it is not possible to supply any input specific parameters to the black-box. To the best of our knowledge, for the first time, we are able to learn input-specific parameters for a black box in this work. As a test application, we choose a popular image denoising method BM3D as our black-box compute. Then, we use a differentiable surrogate model (a neural network) to approximate the black-box behaviour. Next, another neural network is used in an end-to-end fashion to learn input instance-specific parameters for the black-box. Motivated by prior advances in surrogate-based optimization, we applied our method to the Smartphone Image Denoising Dataset (SIDD) and the Color Berkeley Segmentation Dataset (CBSD68) for image denoising. The results are compelling, demonstrating a significant increase in PSNR and a notable improvement in SSIM nearing 0.93. Experimental results underscore the effectiveness of our approach in achieving substantial improvements in both model performance and optimization efficiency. For code and implementation details, please refer to our GitHub repository: https://github.com/arnisha-k/instance-specific-param

Abhineet Singh, Nilanjan Ray

This paper improves upon the Pix2Seq object detector by extending it for videos. In the process, it introduces a new way to perform end-to-end video object detection that improves upon existing video detectors in two key ways. First, by representing objects as variable-length sequences of discrete tokens, we can succinctly represent widely varying numbers of video objects, with diverse shapes and locations, without having to inject any localization cues in the training process. This eliminates the need to sample the space of all possible boxes that constrains conventional detectors and thus solves the dual problems of loss sparsity during training and heuristics-based postprocessing during inference. Second, it conceptualizes and outputs the video objects as fully integrated and indivisible 3D boxes or tracklets instead of generating image-specific 2D boxes and linking these boxes together to construct the video object, as done in most conventional detectors. This allows it to scale effortlessly with available computational resources by simply increasing the length of the video subsequence that the network takes as input, even generalizing to multi-object tracking if the subsequence can span the entire video. We compare our video detector with the baseline Pix2Seq static detector on several datasets and demonstrate consistent improvement, although with strong signs of being bottlenecked by our limited computational resources. We also compare it with several video detectors on UA-DETRAC to show that it is competitive with the current state of the art even with the computational bottleneck. We make our code and models publicly available.

Jakaria Rabbi, Johannes Kiechle, Christian Beaulieu et al.

This paper presents a comprehensive study focused on disentangling hippocampal shape variations from diffusion tensor imaging (DTI) datasets within the context of neurological disorders. Leveraging a Mesh Variational Autoencoder (VAE) enhanced with Supervised Contrastive Learning, our approach aims to improve interpretability by disentangling two distinct latent variables corresponding to age and the presence of diseases. In our ablation study, we investigate a range of VAE architectures and contrastive loss functions, showcasing the enhanced disentanglement capabilities of our approach. This evaluation uses synthetic 3D torus mesh data and real 3D hippocampal mesh datasets derived from the DTI hippocampal dataset. Our supervised disentanglement model outperforms several state-of-the-art (SOTA) methods like attribute and guided VAEs in terms of disentanglement scores. Our model distinguishes between age groups and disease status in patients with Multiple Sclerosis (MS) using the hippocampus data. Our Mesh VAE with Supervised Contrastive Learning shows the volume changes of the hippocampus of MS populations at different ages, and the result is consistent with the current neuroimaging literature. This research provides valuable insights into the relationship between neurological disorder and hippocampal shape changes in different age groups of MS populations using a Mesh VAE with Supervised Contrastive loss. Our code is available at https://github.com/Jakaria08/Explaining_Shape_Variability

Abhineet Singh, Ila Jasra, Omar Mouhammed et al.

This paper presents advancements in automated early-stage prediction of the success of reprogramming human induced pluripotent stem cells (iPSCs) as a potential source for regenerative cell therapies.The minuscule success rate of iPSC-reprogramming of around $ 0.01% $ to $ 0.1% $ makes it labor-intensive, time-consuming, and exorbitantly expensive to generate a stable iPSC line. Since that requires culturing of millions of cells and intense biological scrutiny of multiple clones to identify a single optimal clone. The ability to reliably predict which cells are likely to establish as an optimal iPSC line at an early stage of pluripotency would therefore be ground-breaking in rendering this a practical and cost-effective approach to personalized medicine. Temporal information about changes in cellular appearance over time is crucial for predicting its future growth outcomes. In order to generate this data, we first performed continuous time-lapse imaging of iPSCs in culture using an ultra-high resolution microscope. We then annotated the locations and identities of cells in late-stage images where reliable manual identification is possible. Next, we propagated these labels backwards in time using a semi-automated tracking system to obtain labels for early stages of growth. Finally, we used this data to train deep neural networks to perform automatic cell segmentation and classification. Our code and data are available at https://github.com/abhineet123/ipsc_prediction.

Fateme Bahri, Nilanjan Ray

Background subtraction is a significant task in computer vision and an essential step for many real world applications. One of the challenges for background subtraction methods is dynamic background, which constitute stochastic movements in some parts of the background. In this paper, we have proposed a new background subtraction method, called DBSGen, which uses two generative neural networks, one for dynamic motion removal and another for background generation. At the end, the foreground moving objects are obtained by a pixel-wise distance threshold based on a dynamic entropy map. The proposed method has a unified framework that can be optimized in an end-to-end and unsupervised fashion. The performance of the method is evaluated over dynamic background sequences and it outperforms most of state-of-the-art methods. Our code is publicly available at https://github.com/FatemeBahri/DBSGen.

Soumyadeep Pal, Matthew Tennant, Nilanjan Ray

We present a postprocessing layer for deformable image registration to make a registration field more diffeomorphic by encouraging Jacobians of the transformation to be positive. Diffeomorphic image registration is important for medical imaging studies because of the properties like invertibility, smoothness of the transformation, and topology preservation/non-folding of the grid. Violation of these properties can lead to destruction of the neighbourhood and the connectivity of anatomical structures during image registration. Most of the recent deep learning methods do not explicitly address this folding problem and try to solve it with a smoothness regularization on the registration field. In this paper, we propose a differentiable layer, which takes any registration field as its input, computes exponential of the Jacobian matrices of the input and reconstructs a new registration field from the exponentiated Jacobian matrices using Poisson reconstruction. Our proposed Poisson reconstruction loss enforces positive Jacobians for the final registration field. Thus, our method acts as a post-processing layer without any learnable parameters of its own and can be placed at the end of any deep learning pipeline to form an end-to-end learnable framework. We show the effectiveness of our proposed method for a popular deep learning registration method Voxelmorph and evaluate it with a dataset containing 3D brain MRI scans. Our results show that our post-processing can effectively decrease the number of non-positive Jacobians by a significant amount without any noticeable deterioration of the registration accuracy, thus making the registration field more diffeomorphic. Our code is available online at https://github.com/Soumyadeep-Pal/Diffeomorphic-Image-Registration-Postprocess.

Amir Akbarnejad, Gilbert Bigras, Nilanjan Ray

The analogy between Gaussian processes (GPs) and deep artificial neural networks (ANNs) has received a lot of interest, and has shown promise to unbox the blackbox of deep ANNs. Existing theoretical works put strict assumptions on the ANN (e.g. requiring all intermediate layers to be wide, or using specific activation functions). Accommodating those theoretical assumptions is hard in recent deep architectures, and those theoretical conditions need refinement as new deep architectures emerge. In this paper we derive an evidence lower-bound that encourages the GP's posterior to match the ANN's output without any requirement on the ANN. Using our method we find out that on 5 datasets, only a subset of those theoretical assumptions are sufficient. Indeed, in our experiments we used a normal ResNet-18 or feed-forward backbone with a single wide layer in the end. One limitation of training GPs is the lack of scalability with respect to the number of inducing points. We use novel computational techniques that allow us to train GPs with hundreds of thousands of inducing points and with GPU acceleration. As shown in our experiments, doing so has been essential to get a close match between the GPs and the ANNs on 5 datasets. We implement our method as a publicly available tool called GPEX: https://github.com/amirakbarnejad/gpex. On 5 datasets (4 image datasets, and 1 biological dataset) and ANNs with 2 types of functionality (classifier or attention-mechanism) we were able to find GPs whose outputs closely match those of the corresponding ANNs. After matching the GPs to the ANNs, we used the GPs' kernel functions to explain the ANNs' decisions. We provide more than 200 explanations (around 30 explanations in the paper and the rest in the supplementary) which are highly interpretable by humans and show the ability of the obtained GPs to unbox the ANNs' decisions.

Abhineet Singh, Justin Rozeboom, Nilanjan Ray

This paper presents a new unified approach to semantic segmentation in both images and videos by using language modeling to output the masks as sequences of discrete tokens. We use run length encoding (RLE) to discretize the segmentation masks and then train a modified version of Pix2Seq \cite{p2s} to output these RLE tokens through autoregression. We propose novel tokenization strategies to compress the length of the token sequence to make it practicable to extend this approach to videos. We also show how instance information can be incorporated into the tokenization process to perform panoptic segmentation. We evaluate our proposed models on two datasets to show that they are competitive with the state of the art in spite of being bottlenecked by our limited computational resources.

Ameneh Sheikhjafari, Michelle Noga, Kumaradevan Punithakumar et al.

Diffeomorphic deformable image registration is one of the crucial tasks in medical image analysis, which aims to find a unique transformation while preserving the topology and invertibility of the transformation. Deep convolutional neural networks (CNNs) have yielded well-suited approaches for image registration by learning the transformation priors from a large dataset. The improvement in the performance of these methods is related to their ability to learn information from several sample medical images that are difficult to obtain and bias the framework to the specific domain of data. In this paper, we propose a novel diffeomorphic training-free approach; this is built upon the principle of an ordinary differential equation. Our formulation yields an Euler integration type recursive scheme to estimate the changes of spatial transformations between the fixed and the moving image pyramids at different resolutions. The proposed architecture is simple in design. The moving image is warped successively at each resolution and finally aligned to the fixed image; this procedure is recursive in a way that at each resolution, a fully convolutional network (FCN) models a progressive change of deformation for the current warped image. The entire system is end-to-end and optimized for each pair of images from scratch. In comparison to learning-based methods, the proposed method neither requires a dedicated training set nor suffers from any training bias. We evaluate our method on three cardiac image datasets. The evaluation results demonstrate that the proposed method achieves state-of-the-art registration accuracy while maintaining desirable diffeomorphic properties.

Sara Elkerdawy, Mostafa Elhoushi, Hong Zhang et al.

Dynamic model pruning is a recent direction that allows for the inference of a different sub-network for each input sample during deployment. However, current dynamic methods rely on learning a continuous channel gating through regularization by inducing sparsity loss. This formulation introduces complexity in balancing different losses (e.g task loss, regularization loss). In addition, regularization based methods lack transparent tradeoff hyperparameter selection to realize a computational budget. Our contribution is two-fold: 1) decoupled task and pruning losses. 2) Simple hyperparameter selection that enables FLOPs reduction estimation before training. Inspired by the Hebbian theory in Neuroscience: "neurons that fire together wire together", we propose to predict a mask to process k filters in a layer based on the activation of its previous layer. We pose the problem as a self-supervised binary classification problem. Each mask predictor module is trained to predict if the log-likelihood for each filter in the current layer belongs to the top-k activated filters. The value k is dynamically estimated for each input based on a novel criterion using the mass of heatmaps. We show experiments on several neural architectures, such as VGG, ResNet and MobileNet on CIFAR and ImageNet datasets. On CIFAR, we reach similar accuracy to SOTA methods with 15% and 24% higher FLOPs reduction. Similarly in ImageNet, we achieve lower drop in accuracy with up to 13% improvement in FLOPs reduction.

Ayantha Randika, Nilanjan Ray, Xiao Xiao et al.

Optical character recognition (OCR) is a widely used pattern recognition application in numerous domains. There are several feature-rich, general-purpose OCR solutions available for consumers, which can provide moderate to excellent accuracy levels. However, accuracy can diminish with difficult and uncommon document domains. Preprocessing of document images can be used to minimize the effect of domain shift. In this paper, a novel approach is presented for creating a customized preprocessor for a given OCR engine. Unlike the previous OCR agnostic preprocessing techniques, the proposed approach approximates the gradient of a particular OCR engine to train a preprocessor module. Experiments with two datasets and two OCR engines show that the presented preprocessor is able to improve the accuracy of the OCR up to 46% from the baseline by applying pixel-level manipulations to the document image. The implementation of the proposed method and the enhanced public datasets are available for download.

Logan Gilmour, Nilanjan Ray

CNNs, initially inspired by human vision, differ in a key way: they sample uniformly, rather than with highest density in a focal point. For very large images, this makes training untenable, as the memory and computation required for activation maps scales quadratically with the side length of an image. We propose an image pyramid based approach that extracts narrow glimpses of the of the input image and iteratively refines them to accomplish regression tasks. To assist with high-accuracy regression, we introduce a novel intermediate representation we call 'spatialized features'. Our approach scales logarithmically with the side length, so it works with very large images. We apply our method to Cephalometric X-ray Landmark Detection and get state-of-the-art results.

Abhishek Nan, Matthew Tennant, Uriel Rubin et al.

Autograd-based software packages have recently renewed interest in image registration using homography and other geometric models by gradient descent and optimization, e.g., AirLab and DRMIME. In this work, we emphasize on using complex matrix exponential (CME) over real matrix exponential to compute transformation matrices. CME is theoretically more suitable and practically provides faster convergence as our experiments show. Further, we demonstrate that the use of an ordinary differential equation (ODE) as an optimizable dynamical system can adapt the transformation matrix more accurately to the multi-resolution Gaussian pyramid for image registration. Our experiments include four publicly available benchmark datasets, two of them 2D and the other two being 3D. Experiments demonstrate that our proposed method yields significantly better registration compared to a number of off-the-shelf, popular, state-of-the-art image registration toolboxes.

Sara Elkerdawy, Mostafa Elhoushi, Abhineet Singh et al.

Recent advances in pruning of neural networks have made it possible to remove a large number of filters or weights without any perceptible drop in accuracy. The number of parameters and that of FLOPs are usually the reported metrics to measure the quality of the pruned models. However, the gain in speed for these pruned models is often overlooked in the literature due to the complex nature of latency measurements. In this paper, we show the limitation of filter pruning methods in terms of latency reduction and propose LayerPrune framework. LayerPrune presents a set of layer pruning methods based on different criteria that achieve higher latency reduction than filter pruning methods on similar accuracy. The advantage of layer pruning over filter pruning in terms of latency reduction is a result of the fact that the former is not constrained by the original model's depth and thus allows for a larger range of latency reduction. For each filter pruning method we examined, we use the same filter importance criterion to calculate a per-layer importance score in one-shot. We then prune the least important layers and fine-tune the shallower model which obtains comparable or better accuracy than its filter-based pruning counterpart. This one-shot process allows to remove layers from single path networks like VGG before fine-tuning, unlike in iterative filter pruning, a minimum number of filters per layer is required to allow for data flow which constraint the search space. To the best of our knowledge, we are the first to examine the effect of pruning methods on latency metric instead of FLOPs for multiple networks, datasets and hardware targets. LayerPrune also outperforms handcrafted architectures such as Shufflenet, MobileNet, MNASNet and ResNet18 by 7.3%, 4.6%, 2.8% and 0.5% respectively on similar latency budget on ImageNet dataset.

Jakaria Rabbi, Nilanjan Ray, Matthias Schubert et al.

The detection performance of small objects in remote sensing images is not satisfactory compared to large objects, especially in low-resolution and noisy images. A generative adversarial network (GAN)-based model called enhanced super-resolution GAN (ESRGAN) shows remarkable image enhancement performance, but reconstructed images miss high-frequency edge information. Therefore, object detection performance degrades for small objects on recovered noisy and low-resolution remote sensing images. Inspired by the success of edge enhanced GAN (EEGAN) and ESRGAN, we apply a new edge-enhanced super-resolution GAN (EESRGAN) to improve the image quality of remote sensing images and use different detector networks in an end-to-end manner where detector loss is backpropagated into the EESRGAN to improve the detection performance. We propose an architecture with three components: ESRGAN, Edge Enhancement Network (EEN), and Detection network. We use residual-in-residual dense blocks (RRDB) for both the ESRGAN and EEN, and for the detector network, we use the faster region-based convolutional network (FRCNN) (two-stage detector) and single-shot multi-box detector (SSD) (one stage detector). Extensive experiments on a public (car overhead with context) and a self-assembled (oil and gas storage tank) satellite dataset show superior performance of our method compared to the standalone state-of-the-art object detectors.

Abhishek Nan, Matthew Tennant, Uriel Rubin et al.

In this work, we present a novel unsupervised image registration algorithm. It is differentiable end-to-end and can be used for both multi-modal and mono-modal registration. This is done using mutual information (MI) as a metric. The novelty here is that rather than using traditional ways of approximating MI, we use a neural estimator called MINE and supplement it with matrix exponential for transformation matrix computation. This leads to improved results as compared to the standard algorithms available out-of-the-box in state-of-the-art image registration toolboxes.

Abhineet Singh, Marcin Pietrasik, Gabriell Natha et al.

Automatic detection of animals that have strayed into human inhabited areas has important security and road safety applications. This paper attempts to solve this problem using deep learning techniques from a variety of computer vision fields including object detection, tracking, segmentation and edge detection. Several interesting insights into transfer learning are elicited while adapting models trained on benchmark datasets for real world deployment. Empirical evidence is presented to demonstrate the inability of detectors to generalize from training images of animals in their natural habitats to deployment scenarios of man-made environments. A solution is also proposed using semi-automated synthetic data generation for domain specific training. Code and data used in the experiments are made available to facilitate further work in this domain.

Sara Elkerdawy, Hong Zhang, Nilanjan Ray

Convolutional neural networks (CNNs) have emerged as the state-of-the-art in multiple vision tasks including depth estimation. However, memory and computing power requirements remain as challenges to be tackled in these models. Monocular depth estimation has significant use in robotics and virtual reality that requires deployment on low-end devices. Training a small model from scratch results in a significant drop in accuracy and it does not benefit from pre-trained large models. Motivated by the literature of model pruning, we propose a lightweight monocular depth model obtained from a large trained model. This is achieved by removing the least important features with a novel joint end-to-end filter pruning. We propose to learn a binary mask for each filter to decide whether to drop the filter or not. These masks are trained jointly to exploit relations between filters at different layers as well as redundancy within the same layer. We show that we can achieve around 5x compression rate with small drop in accuracy on the KITTI driving dataset. We also show that masking can improve accuracy over the baseline with fewer parameters, even without enforcing compression loss.

Abhineet Singh, Hayden Kalke, Mark Loewen et al.

This paper deals with the problem of computing surface ice concentration for two different types of ice from digital images of river surface. It presents the results of attempting to solve this problem using several state of the art semantic segmentation methods based on deep convolutional neural networks (CNNs). This task presents two main challenges - very limited availability of labeled training data and presence of noisy labels due to the great difficulty of visually distinguishing between the two types of ice, even for human experts. The results are used to analyze the extent to which some of the best deep learning methods currently in existence can handle these challenges. The code and data used in the experiments are made publicly available to facilitate further work in this domain.

Nhat M. Nguyen, Nilanjan Ray

Differentiable programming is able to combine different functions or programs in a processing pipeline with the goal of applying end-to-end learning or optimization. A significant impediment is the non-differentiable nature of some algorithms. We propose to use synthetic gradients (SG) to overcome this difficulty. SG uses the universal function approximation property of neural networks. We apply SG to combine convolutional neural network (CNN) with dynamic programming (DP) in end-to-end learning for segmenting left ventricle from short axis view of heart MRI. Our experiments show that end-to-end combination of CNN and DP requires fewer labeled images to achieve a significantly better segmentation accuracy than using only CNN.

Yao Xue, Gilbert Bigras, Judith Hugh et al.

Automated cell detection and localization from microscopy images are significant tasks in biomedical research and clinical practice. In this paper, we design a new cell detection and localization algorithm that combines deep convolutional neural network (CNN) and compressed sensing (CS) or sparse coding (SC) for end-to-end training. We also derive, for the first time, a backpropagation rule, which is applicable to train any algorithm that implements a sparse code recovery layer. The key observation behind our algorithm is that cell detection task is a point object detection task in computer vision, where the cell centers (i.e., point objects) occupy only a tiny fraction of the total number of pixels in an image. Thus, we can apply compressed sensing (or, equivalently sparse coding) to compactly represent a variable number of cells in a projected space. Then, CNN regresses this compressed vector from the input microscopy image. Thanks to the SC/CS recovery algorithm (L1 optimization) that can recover sparse cell locations from the output of CNN. We train this entire processing pipeline end-to-end and demonstrate that end-to-end training provides accuracy improvements over a training paradigm that treats CNN and CS-recovery layers separately. Our algorithm design also takes into account a form of ensemble average of trained models naturally to further boost accuracy of cell detection. We have validated our algorithm on benchmark datasets and achieved excellent performances.

Fateme Bahri, Moein Shakeri, Nilanjan Ray

Moving object detection (MOD) is a significant problem in computer vision that has many real world applications. Different categories of methods have been proposed to solve MOD. One of the challenges is to separate moving objects from illumination changes and shadows that are present in most real world videos. State-of-the-art methods that can handle illumination changes and shadows work in a batch mode; thus, these methods are not suitable for long video sequences or real-time applications. In this paper, we propose an extension of a state-of-the-art batch MOD method (ILISD) to an online/incremental MOD using unsupervised and generative neural networks, which use illumination invariant image representations. For each image in a sequence, we use a low-dimensional representation of a background image by a neural network and then based on the illumination invariant representation, decompose the foreground image into: illumination change and moving objects. Optimization is performed by stochastic gradient descent in an end-to-end and unsupervised fashion. Our algorithm can work in both batch and online modes. In the batch mode, like other batch methods, optimizer uses all the images. In online mode, images can be incrementally fed into the optimizer. Based on our experimental evaluation on benchmark image sequences, both the online and the batch modes of our algorithm achieve state-of-the-art accuracy on most data sets.

Nhat M. Nguyen, Nilanjan Ray

Most existing GANs architectures that generate images use transposed convolution or resize-convolution as their upsampling algorithm from lower to higher resolution feature maps in the generator. We argue that this kind of fixed operation is problematic for GANs to model objects that have very different visual appearances. We propose a novel adaptive convolution method that learns the upsampling algorithm based on the local context at each location to address this problem. We modify a baseline GANs architecture by replacing normal convolutions with adaptive convolutions in the generator. Experiments on CIFAR-10 dataset show that our modified models improve the baseline model by a large margin. Furthermore, our models achieve state-of-the-art performance on CIFAR-10 and STL-10 datasets in the unsupervised setting.

Yao Xue, Nilanjan Ray

The ability to automatically detect certain types of cells or cellular subunits in microscopy images is of significant interest to a wide range of biomedical research and clinical practices. Cell detection methods have evolved from employing hand-crafted features to deep learning-based techniques. The essential idea of these methods is that their cell classifiers or detectors are trained in the pixel space, where the locations of target cells are labeled. In this paper, we seek a different route and propose a convolutional neural network (CNN)-based cell detection method that uses encoding of the output pixel space. For the cell detection problem, the output space is the sparsely labeled pixel locations indicating cell centers. We employ random projections to encode the output space to a compressed vector of fixed dimension. Then, CNN regresses this compressed vector from the input pixels. Furthermore, it is possible to stably recover sparse cell locations on the output pixel space from the predicted compressed vector using $L_1$-norm optimization. In the past, output space encoding using compressed sensing (CS) has been used in conjunction with linear and non-linear predictors. To the best of our knowledge, this is the first successful use of CNN with CS-based output space encoding. We made substantial experiments on several benchmark datasets, where the proposed CNN + CS framework (referred to as CNNCS) achieved the highest or at least top-3 performance in terms of F1-score, compared with other state-of-the-art methods.

Homa Foroughi, Moein Shakeri, Nilanjan Ray et al.

Face recognition has been widely studied due to its importance in different applications; however, most of the proposed methods fail when face images are occluded or captured under illumination and pose variations. Recently several low-rank dictionary learning methods have been proposed and achieved promising results for noisy observations. While these methods are mostly developed for single-modality scenarios, recent studies demonstrated the advantages of feature fusion from multiple inputs. We propose a multi-modal structured low-rank dictionary learning method for robust face recognition, using raw pixels of face images and their illumination invariant representation. The proposed method learns robust and discriminative representations from contaminated face images, even if there are few training samples with large intra-class variations. Extensive experiments on different datasets validate the superior performance and robustness of our method to severe illumination variations and occlusion.

Homa Foroughi, Nilanjan Ray, Hong Zhang

For an object classification system, the most critical obstacles towards real-world applications are often caused by large intra-class variability, arising from different lightings, occlusion and corruption, in limited sample sets. Most methods in the literature would fail when the training samples are heavily occluded, corrupted or have significant illumination or viewpoint variations. Besides, most of the existing methods and especially deep learning-based methods, need large training sets to achieve a satisfactory recognition performance. Although using the pre-trained network on a generic large-scale dataset and fine-tune it to the small-sized target dataset is a widely used technique, this would not help when the content of base and target datasets are very different. To address these issues, we propose a joint projection and low-rank dictionary learning method using dual graph constraints (JP-LRDL). The proposed joint learning method would enable us to learn the features on top of which dictionaries can be better learned, from the data with large intra-class variability. Specifically, a structured class-specific dictionary is learned and the discrimination is further improved by imposing a graph constraint on the coding coefficients, that maximizes the intra-class compactness and inter-class separability. We also enforce low-rank and structural incoherence constraints on sub-dictionaries to make them more compact and robust to variations and outliers and reduce the redundancy among them, respectively. To preserve the intrinsic structure of data and penalize unfavourable relationship among training samples simultaneously, we introduce a projection graph into the framework, which significantly enhances the discriminative ability of the projection matrix and makes the method robust to small-sized and high-dimensional datasets.

Sayan Ghosal, Nilanjan Ray

Deformable registration is ubiquitous in medical image analysis. Many deformable registration methods minimize sum of squared difference (SSD) as the registration cost with respect to deformable model parameters. In this work, we construct a tight upper bound of the SSD registration cost by using a fully convolutional neural network (FCNN) in the registration pipeline. The upper bound SSD (UB-SSD) enhances the original deformable model parameter space by adding a heatmap output from FCNN. Next, we minimize this UB-SSD by adjusting both the parameters of the FCNN and the parameters of the deformable model in coordinate descent. Our coordinate descent framework is end-to-end and can work with any deformable registration method that uses SSD. We demonstrate experimentally that our method enhances the accuracy of deformable registration algorithms significantly on two publicly available 3D brain MRI data sets.

Mennatullah Siam, Sepehr Valipour, Martin Jagersand et al.

Semantic segmentation has recently witnessed major progress, where fully convolutional neural networks have shown to perform well. However, most of the previous work focused on improving single image segmentation. To our knowledge, no prior work has made use of temporal video information in a recurrent network. In this paper, we introduce a novel approach to implicitly utilize temporal data in videos for online semantic segmentation. The method relies on a fully convolutional network that is embedded into a gated recurrent architecture. This design receives a sequence of consecutive video frames and outputs the segmentation of the last frame. Convolutional gated recurrent networks are used for the recurrent part to preserve spatial connectivities in the image. Our proposed method can be applied in both online and batch segmentation. This architecture is tested for both binary and semantic video segmentation tasks. Experiments are conducted on the recent benchmarks in SegTrack V2, Davis, CityScapes, and Synthia. Using recurrent fully convolutional networks improved the baseline network performance in all of our experiments. Namely, 5% and 3% improvement of F-measure in SegTrack2 and Davis respectively, 5.7% improvement in mean IoU in Synthia and 3.5% improvement in categorical mean IoU in CityScapes. The performance of the RFCN network depends on its baseline fully convolutional network. Thus RFCN architecture can be seen as a method to improve its baseline segmentation network by exploiting spatiotemporal information in videos.

Sepehr Valipour, Mennatullah Siam, Martin Jagersand et al.

Image segmentation is an important step in most visual tasks. While convolutional neural networks have shown to perform well on single image segmentation, to our knowledge, no study has been been done on leveraging recurrent gated architectures for video segmentation. Accordingly, we propose a novel method for online segmentation of video sequences that incorporates temporal data. The network is built from fully convolutional element and recurrent unit that works on a sliding window over the temporal data. We also introduce a novel convolutional gated recurrent unit that preserves the spatial information and reduces the parameters learned. Our method has the advantage that it can work in an online fashion instead of operating over the whole input batch of video frames. The network is tested on the change detection dataset, and proved to have 5.5\% improvement in F-measure over a plain fully convolutional network for per frame segmentation. It was also shown to have improvement of 1.4\% for the F-measure compared to our baseline network that we call FCN 12s.

Homa Foroughi, Nilanjan Ray, Hong Zhang

In this paper, we aim at learning simultaneously a discriminative dictionary and a robust projection matrix from noisy data. The joint learning, makes the learned projection and dictionary a better fit for each other, so a more accurate classification can be obtained. However, current prevailing joint dimensionality reduction and dictionary learning methods, would fail when the training samples are noisy or heavily corrupted. To address this issue, we propose a joint projection and dictionary learning using low-rank regularization and graph constraints (JPDL-LR). Specifically, the discrimination of the dictionary is achieved by imposing Fisher criterion on the coding coefficients. In addition, our method explicitly encodes the local structure of data by incorporating a graph regularization term, that further improves the discriminative ability of the projection matrix. Inspired by recent advances of low-rank representation for removing outliers and noise, we enforce a low-rank constraint on sub-dictionaries of all classes to make them more compact and robust to noise. Experimental results on several benchmark datasets verify the effectiveness and robustness of our method for both dimensionality reduction and image classification, especially when the data contains considerable noise or variations.