Yash Patel

Filip Radenovic, Abhimanyu Dubey, Abhishek Kadian et al. · meta-ai

Vision-language models trained with contrastive learning on large-scale noisy data are becoming increasingly popular for zero-shot recognition problems. In this paper we improve the following three aspects of the contrastive pre-training pipeline: dataset noise, model initialization and the training objective. First, we propose a straightforward filtering strategy titled Complexity, Action, and Text-spotting (CAT) that significantly reduces dataset size, while achieving improved performance across zero-shot vision-language tasks. Next, we propose an approach titled Concept Distillation to leverage strong unimodal representations for contrastive training that does not increase training complexity while outperforming prior work. Finally, we modify the traditional contrastive alignment objective, and propose an importance-sampling approach to up-sample the importance of hard-negatives without adding additional complexity. On an extensive zero-shot benchmark of 29 tasks, our Distilled and Hard-negative Training (DiHT) approach improves on 20 tasks compared to the baseline. Furthermore, for few-shot linear probing, we propose a novel approach that bridges the gap between zero-shot and few-shot performance, substantially improving over prior work. Models are available at https://github.com/facebookresearch/diht.

Štěpán Šimsa, Milan Šulc, Michal Uřičář et al.

This paper introduces the DocILE benchmark with the largest dataset of business documents for the tasks of Key Information Localization and Extraction and Line Item Recognition. It contains 6.7k annotated business documents, 100k synthetically generated documents, and nearly~1M unlabeled documents for unsupervised pre-training. The dataset has been built with knowledge of domain- and task-specific aspects, resulting in the following key features: (i) annotations in 55 classes, which surpasses the granularity of previously published key information extraction datasets by a large margin; (ii) Line Item Recognition represents a highly practical information extraction task, where key information has to be assigned to items in a table; (iii) documents come from numerous layouts and the test set includes zero- and few-shot cases as well as layouts commonly seen in the training set. The benchmark comes with several baselines, including RoBERTa, LayoutLMv3 and DETR-based Table Transformer; applied to both tasks of the DocILE benchmark, with results shared in this paper, offering a quick starting point for future work. The dataset, baselines and supplementary material are available at https://github.com/rossumai/docile.

Amazon AGI, Aaron Langford, Aayush Shah et al. · amazon-science

We present Amazon Nova, a new generation of state-of-the-art foundation models that deliver frontier intelligence and industry-leading price performance. Amazon Nova Pro is a highly-capable multimodal model with the best combination of accuracy, speed, and cost for a wide range of tasks. Amazon Nova Lite is a low-cost multimodal model that is lightning fast for processing images, video, documents and text. Amazon Nova Micro is a text-only model that delivers our lowest-latency responses at very low cost. Amazon Nova Canvas is an image generation model that creates professional grade images with rich customization controls. Amazon Nova Reel is a video generation model offering high-quality outputs, customization, and motion control. Our models were built responsibly and with a commitment to customer trust, security, and reliability. We report benchmarking results for core capabilities, agentic performance, long context, functional adaptation, runtime performance, and human evaluation.

Tong Wei, Yash Patel, Alexander Shekhovtsov et al.

We propose $\nabla$-RANSAC, a generalized differentiable RANSAC that allows learning the entire randomized robust estimation pipeline. The proposed approach enables the use of relaxation techniques for estimating the gradients in the sampling distribution, which are then propagated through a differentiable solver. The trainable quality function marginalizes over the scores from all the models estimated within $\nabla$-RANSAC to guide the network learning accurate and useful inlier probabilities or to train feature detection and matching networks. Our method directly maximizes the probability of drawing a good hypothesis, allowing us to learn better sampling distributions. We test $\nabla$-RANSAC on various real-world scenarios on fundamental and essential matrix estimation, and 3D point cloud registration, outdoors and indoors, with handcrafted and learning-based features. It is superior to the state-of-the-art in terms of accuracy while running at a similar speed to its less accurate alternatives. The code and trained models are available at https://github.com/weitong8591/differentiable_ransac.

Yash Patel, Yusheng Xie, Yi Zhu et al. · amazon-science

Learning to segment images purely by relying on the image-text alignment from web data can lead to sub-optimal performance due to noise in the data. The noise comes from the samples where the associated text does not correlate with the image's visual content. Instead of purely relying on the alignment from the noisy data, this paper proposes a novel loss function termed SimCon, which accounts for intra-modal similarities to determine the appropriate set of positive samples to align. Further, using multiple views of the image (created synthetically) for training and combining the SimCon loss with it makes the training more robust. This version of the loss is termed MV-SimCon. The empirical results demonstrate that using the proposed loss function leads to consistent improvements on zero-shot, text supervised semantic segmentation and outperforms state-of-the-art by $+3.0\%$, $+3.3\%$ and $+6.9\%$ on PASCAL VOC, PASCAL Context and MSCOCO, respectively. With test time augmentations, we set a new record by improving these results further to $58.7\%$, $26.6\%$, and $33.3\%$ on PASCAL VOC, PASCAL Context, and MSCOCO, respectively. In addition, using the proposed loss function leads to robust training and faster convergence.

Štěpán Šimsa, Milan Šulc, Matyáš Skalický et al.

The lack of data for information extraction (IE) from semi-structured business documents is a real problem for the IE community. Publications relying on large-scale datasets use only proprietary, unpublished data due to the sensitive nature of such documents. Publicly available datasets are mostly small and domain-specific. The absence of a large-scale public dataset or benchmark hinders the reproducibility and cross-evaluation of published methods. The DocILE 2023 competition, hosted as a lab at the CLEF 2023 conference and as an ICDAR 2023 competition, will run the first major benchmark for the tasks of Key Information Localization and Extraction (KILE) and Line Item Recognition (LIR) from business documents. With thousands of annotated real documents from open sources, a hundred thousand of generated synthetic documents, and nearly a million unlabeled documents, the DocILE lab comes with the largest publicly available dataset for KILE and LIR. We are looking forward to contributions from the Computer Vision, Natural Language Processing, Information Retrieval, and other communities. The data, baselines, code and up-to-date information about the lab and competition are available at https://docile.rossum.ai/.

Yash Patel, Tirth Shah, Mrinal Kanti Dhar et al.

The global burden of acute and chronic wounds presents a compelling case for enhancing wound classification methods, a vital step in diagnosing and determining optimal treatments. Recognizing this need, we introduce an innovative multi-modal network based on a deep convolutional neural network for categorizing wounds into four categories: diabetic, pressure, surgical, and venous ulcers. Our multi-modal network uses wound images and their corresponding body locations for more precise classification. A unique aspect of our methodology is incorporating a body map system that facilitates accurate wound location tagging, improving upon traditional wound image classification techniques. A distinctive feature of our approach is the integration of models such as VGG16, ResNet152, and EfficientNet within a novel architecture. This architecture includes elements like spatial and channel-wise Squeeze-and-Excitation modules, Axial Attention, and an Adaptive Gated Multi-Layer Perceptron, providing a robust foundation for classification. Our multi-modal network was trained and evaluated on two distinct datasets comprising relevant images and corresponding location information. Notably, our proposed network outperformed traditional methods, reaching an accuracy range of 74.79% to 100% for Region of Interest (ROI) without location classifications, 73.98% to 100% for ROI with location classifications, and 78.10% to 100% for whole image classifications. This marks a significant enhancement over previously reported performance metrics in the literature. Our results indicate the potential of our multi-modal network as an effective decision-support tool for wound image classification, paving the way for its application in various clinical contexts.

Hosein Barzekar, Yash Patel, Ling Tong et al.

Cancer is a leading cause of death in many countries. An early diagnosis of cancer based on biomedical imaging ensures effective treatment and a better prognosis. However, biomedical imaging presents challenges to both clinical institutions and researchers. Physiological anomalies are often characterized by slight abnormalities in individual cells or tissues, making them difficult to detect visually. Traditionally, anomalies are diagnosed by radiologists and pathologists with extensive training. This procedure, however, demands the participation of professionals and incurs a substantial cost. The cost makes large-scale biological image classification impractical. In this study, we provide unique deep neural network designs for multiclass classification of medical images, in particular cancer images. We incorporated transformers into a multiclass framework to take advantage of data-gathering capability and perform more accurate classifications. We evaluated models on publicly accessible datasets using various measures to ensure the reliability of the models. Extensive assessment metrics suggest this method can be used for a multitude of classification tasks.

D. M. Anisuzzaman, Yash Patel, Jeffrey Niezgoda et al.

The classification of wound severity is a critical step in wound diagnosis. An effective classifier can help wound professionals categorize wound conditions more quickly and affordably, allowing them to choose the best treatment option. This study used wound photos to construct a deep neural network-based wound severity classifier that classified them into one of three classes: green, yellow, or red. The green class denotes wounds still in the early stages of healing and are most likely to recover with adequate care. Wounds in the yellow category require more attention and treatment than those in the green category. Finally, the red class denotes the most severe wounds that require prompt attention and treatment. A dataset containing different types of wound images is designed with the help of wound specialists. Nine deep learning models are used with applying the concept of transfer learning. Several stacked models are also developed by concatenating these transfer learning models. The maximum accuracy achieved on multi-class classification is 68.49%. In addition, we achieved 78.79%, 81.40%, and 77.57% accuracies on green vs. yellow, green vs. red, and yellow vs. red classifications for binary classifications.

Rahaf Aljundi, Yash Patel, Milan Sulc et al.

Cross entropy loss has served as the main objective function for classification-based tasks. Widely deployed for learning neural network classifiers, it shows both effectiveness and a probabilistic interpretation. Recently, after the success of self supervised contrastive representation learning methods, supervised contrastive methods have been proposed to learn representations and have shown superior and more robust performance, compared to solely training with cross entropy loss. However, cross entropy loss is still needed to train the final classification layer. In this work, we investigate the possibility of learning both the representation and the classifier using one objective function that combines the robustness of contrastive learning and the probabilistic interpretation of cross entropy loss. First, we revisit a previously proposed contrastive-based objective function that approximates cross entropy loss and present a simple extension to learn the classifier jointly. Second, we propose a new version of the supervised contrastive training that learns jointly the parameters of the classifier and the backbone of the network. We empirically show that our proposed objective functions show a significant improvement over the standard cross entropy loss with more training stability and robustness in various challenging settings.

Zhiwei Xue, Yuhang Li, Yash Patel et al.

Telescopes capture images with a particular point spread function (PSF). Inferring what an image would have looked like with a much sharper PSF, a problem known as PSF deconvolution, is ill-posed because PSF convolution is not an invertible transformation. Deep generative models are appealing for PSF deconvolution because they can infer a posterior distribution over candidate images that, if convolved with the PSF, could have generated the observation. However, classical deep generative models such as VAEs and GANs often provide inadequate sample diversity. As an alternative, we propose a classifier-free conditional diffusion model for PSF deconvolution of galaxy images. We demonstrate that this diffusion model captures a greater diversity of possible deconvolutions compared to a conditional VAE.

Yash Patel, Sahana Rayan, Ambuj Tewari

Data-driven approaches to predict-then-optimize decision-making problems seek to mitigate the risk of uncertainty region misspecification in safety-critical settings. Current approaches, however, suffer from considering overly conservative uncertainty regions, often resulting in suboptimal decisionmaking. To this end, we propose Conformal-Predict-Then-Optimize (CPO), a framework for leveraging highly informative, nonconvex conformal prediction regions over high-dimensional spaces based on conditional generative models, which have the desired distribution-free coverage guarantees. Despite guaranteeing robustness, such black-box optimization procedures alone inspire little confidence owing to the lack of explanation of why a particular decision was found to be optimal. We, therefore, augment CPO to additionally provide semantically meaningful visual summaries of the uncertainty regions to give qualitative intuition for the optimal decision. We highlight the CPO framework by demonstrating results on a suite of simulation-based inference benchmark tasks and a vehicle routing task based on probabilistic weather prediction.

Yash Patel, Ambuj Tewari

The solution of PDEs in decision-making tasks is increasingly being undertaken with the help of neural operator surrogate models due to the need for repeated evaluation. Such methods, while significantly more computationally favorable compared to their numerical counterparts, fail to provide any calibrated notions of uncertainty in their predictions. Current methods approach this deficiency typically with ensembling or Bayesian posterior estimation. However, these approaches either require distributional assumptions that fail to hold in practice or lack practical scalability, limiting their applications in practice. We, therefore, propose a novel application of conformal prediction to produce distribution-free uncertainty quantification over the function spaces mapped by neural operators. We then demonstrate how such prediction regions enable a formal regret characterization if leveraged in downstream robust decision-making tasks. We further demonstrate how such posited robust decision-making tasks can be efficiently solved using an infinite-dimensional generalization of Danskin's Theorem and calculus of variations and empirically demonstrate the superior performance of our proposed method over more restrictive modeling paradigms, such as Gaussian Processes, across several engineering tasks.

Mrinal Kanti Dhar, Chuanbo Wang, Yash Patel et al.

Identifying individual tissues, so-called tissue segmentation, in diabetic foot ulcer (DFU) images is a challenging task and little work has been published, largely due to the limited availability of a clinical image dataset. To address this gap, we have created a DFUTissue dataset for the research community to evaluate wound tissue segmentation algorithms. The dataset contains 110 images with tissues labeled by wound experts and 600 unlabeled images. Additionally, we conducted a pilot study on segmenting wound characteristics including fibrin, granulation, and callus using deep learning. Due to the limited amount of annotated data, our framework consists of both supervised learning (SL) and semi-supervised learning (SSL) phases. In the SL phase, we propose a hybrid model featuring a Mix Transformer (MiT-b3) in the encoder and a CNN in the decoder, enhanced by the integration of a parallel spatial and channel squeeze-and-excitation (P-scSE) module known for its efficacy in improving boundary accuracy. The SSL phase employs a pseudo-labeling-based approach, iteratively identifying and incorporating valuable unlabeled images to enhance overall segmentation performance. Comparative evaluations with state-of-the-art methods are conducted for both SL and SSL phases. The SL achieves a Dice Similarity Coefficient (DSC) of 84.89%, which has been improved to 87.64% in the SSL phase. Furthermore, the results are benchmarked against two widely used SSL approaches: Generative Adversarial Networks and Cross-Consistency Training. Additionally, our hybrid model outperforms the state-of-the-art methods with a 92.99% DSC in performing binary segmentation of DFU wound areas when tested on the Chronic Wound dataset. Codes and data are available at https://github.com/uwm-bigdata/DFUTissueSegNet.

Mrinal Kanti Dhar, Taiyu Zhang, Yash Patel et al.

This paper presents FUSegNet, a new model for foot ulcer segmentation in diabetes patients, which uses the pre-trained EfficientNet-b7 as a backbone to address the issue of limited training samples. A modified spatial and channel squeeze-and-excitation (scSE) module called parallel scSE or P-scSE is proposed that combines additive and max-out scSE. A new arrangement is introduced for the module by fusing it in the middle of each decoder stage. As the top decoder stage carries a limited number of feature maps, max-out scSE is bypassed there to form a shorted P-scSE. A set of augmentations, comprising geometric, morphological, and intensity-based augmentations, is applied before feeding the data into the network. The proposed model is first evaluated on a publicly available chronic wound dataset where it achieves a data-based dice score of 92.70%, which is the highest score among the reported approaches. The model outperforms other scSE-based UNet models in terms of Pratt's figure of merits (PFOM) scores in most categories, which evaluates the accuracy of edge localization. The model is then tested in the MICCAI 2021 FUSeg challenge, where a variation of FUSegNet called x-FUSegNet is submitted. The x-FUSegNet model, which takes the average of outputs obtained by FUSegNet using 5-fold cross-validation, achieves a dice score of 89.23%, placing it at the top of the FUSeg Challenge leaderboard. The source code for the model is available on https://github.com/mrinal054/FUSegNet.

Deepak Kumar, Divakar Yadav, Yash Patel

We present a single-GPU (H100, bf16) evaluation of GPT-OSS-20B (Mixture-of-Experts; 20.9B total, approx. 3.61B active) against dense baselines Qwen3-32B and Yi-34B across multiple dimensions. We measure true time-to-first-token (TTFT), full-decode throughput (TPOT), end-to-end latency percentiles, peak VRAM with past key values (PKV) held, and energy via a consistent nvidia-smi-based sampler. At a 2048-token context with 64-token decode, GPT-OSS-20B delivers higher decode throughput and tokens per Joule than dense baselines Qwen3-32B and Yi-34B, while substantially reducing peak VRAM and energy per 1000 generated tokens; its TTFT is higher due to MoE routing overhead. With only 17.3% of parameters active (3.61B of 20.9B), GPT-OSS-20B provides about 31.8% higher decode throughput and 25.8% lower energy per 1000 generated tokens than Qwen3-32B at 2048/64, while using 31.7% less peak VRAM. Normalized by active parameters, GPT-OSS-20B shows markedly stronger per-active-parameter efficiency (APE), underscoring MoE's deployment advantages. We do not evaluate accuracy; this is a deployment-focused study. We release code and consolidated results to enable replication and extension.

Sahana Rayan, Yash Patel, Ambuj Tewari

When solving PDEs, classical numerical solvers are often computationally expensive, while machine learning methods can suffer from spectral bias, failing to capture high-frequency components. Designing an optimal hybrid iterative solver--where, at each iteration, a solver is selected from an ensemble of solvers to leverage their complementary strengths--poses a challenging combinatorial problem. While the greedy selection strategy is desirable for its constant-factor approximation guarantee to the optimal solution, it requires knowledge of the true error at each step, which is generally unavailable in practice. We address this by proposing an approximate greedy router that efficiently mimics a greedy approach to solver selection. Empirical results on the Poisson and Helmholtz equations demonstrate that our method outperforms single-solver baselines and existing hybrid solver approaches, such as HINTS, achieving faster and more stable convergence.

Prarabdh Shukla, Wei Yin Chong, Yash Patel et al.

To meet the demands of content moderation, online platforms have resorted to automated systems. Newer forms of real-time engagement($\textit{e.g.}$, users commenting on live streams) on platforms like Twitch exert additional pressures on the latency expected of such moderation systems. Despite their prevalence, relatively little is known about the effectiveness of these systems. In this paper, we conduct an audit of Twitch's automated moderation tool ($\texttt{AutoMod}$) to investigate its effectiveness in flagging hateful content. For our audit, we create streaming accounts to act as siloed test beds, and interface with the live chat using Twitch's APIs to send over $107,000$ comments collated from $4$ datasets. We measure $\texttt{AutoMod}$'s accuracy in flagging blatantly hateful content containing misogyny, racism, ableism and homophobia. Our experiments reveal that a large fraction of hateful messages, up to $94\%$ on some datasets, $\textit{bypass moderation}$. Contextual addition of slurs to these messages results in $100\%$ removal, revealing $\texttt{AutoMod}$'s reliance on slurs as a moderation signal. We also find that contrary to Twitch's community guidelines, $\texttt{AutoMod}$ blocks up to $89.5\%$ of benign examples that use sensitive words in pedagogical or empowering contexts. Overall, our audit points to large gaps in $\texttt{AutoMod}$'s capabilities and underscores the importance for such systems to understand context effectively.

Yash Patel, Unique Subedi, Ambuj Tewari

We consider the problem of learning the evolution operator for the time-dependent Schrödinger equation, where the Hamiltonian may vary with time. Existing neural network-based surrogates often ignore fundamental properties of the Schrödinger equation, such as linearity and unitarity, and lack theoretical guarantees on prediction error or time generalization. To address this, we introduce a linear estimator for the evolution operator that preserves a weak form of unitarity. We establish both upper and lower bounds on the prediction error that hold uniformly over all sufficiently smooth initial wave functions. Additionally, we derive time generalization bounds that quantify how the estimator extrapolates beyond the time points seen during training. Experiments across real-world Hamiltonians -- including hydrogen atoms, ion traps for qubit design, and optical lattices -- show that our estimator achieves relative errors $10^{-2}$ to $10^{-3}$ times smaller than state-of-the-art methods such as the Fourier Neural Operator and DeepONet.

Abhiti Mishra, Yash Patel, Ambuj Tewari

Transformers robustly exhibit the ability to perform in-context learning, whereby their predictive accuracy on a task can increase not by parameter updates but merely with the placement of training samples in their context windows. Recent works have shown that transformers achieve this by implementing gradient descent in their forward passes. Such results, however, are restricted to standard transformer architectures, which handle finite-dimensional inputs. In the space of PDE surrogate modeling, a generalization of transformers to handle infinite-dimensional function inputs, known as "continuum transformers," has been proposed and similarly observed to exhibit in-context learning. Despite impressive empirical performance, such in-context learning has yet to be theoretically characterized. We herein demonstrate that continuum transformers perform in-context operator learning by performing gradient descent in an operator RKHS. We demonstrate this using novel proof strategies that leverage a generalized representer theorem for Hilbert spaces and gradient flows over the space of functionals of a Hilbert space. We additionally show the operator learned in context is the Bayes Optimal Predictor in the infinite depth limit of the transformer. We then provide empirical validations of this optimality result and demonstrate that the parameters under which such gradient descent is performed are recovered through the continuum transformer training.

Yash Patel, Giorgos Tolias, Jiri Matas

This paper addresses supervised deep metric learning for open-set image retrieval, focusing on three key aspects: the loss function, mixup regularization, and model initialization. In deep metric learning, optimizing the retrieval evaluation metric, recall@k, via gradient descent is desirable but challenging due to its non-differentiable nature. To overcome this, we propose a differentiable surrogate loss that is computed on large batches, nearly equivalent to the entire training set. This computationally intensive process is made feasible through an implementation that bypasses the GPU memory limitations. Additionally, we introduce an efficient mixup regularization technique that operates on pairwise scalar similarities, effectively increasing the batch size even further. The training process is further enhanced by initializing the vision encoder using foundational models, which are pre-trained on large-scale datasets. Through a systematic study of these components, we demonstrate that their synergy enables large models to nearly solve popular benchmarks.

Yash Patel, Declan McNamara, Jackson Loper et al.

Amortized variational inference is an often employed framework in simulation-based inference that produces a posterior approximation that can be rapidly computed given any new observation. Unfortunately, there are few guarantees about the quality of these approximate posteriors. We propose Conformalized Amortized Neural Variational Inference (CANVI), a procedure that is scalable, easily implemented, and provides guaranteed marginal coverage. Given a collection of candidate amortized posterior approximators, CANVI constructs conformalized predictors based on each candidate, compares the predictors using a metric known as predictive efficiency, and returns the most efficient predictor. CANVI ensures that the resulting predictor constructs regions that contain the truth with a user-specified level of probability. CANVI is agnostic to design decisions in formulating the candidate approximators and only requires access to samples from the forward model, permitting its use in likelihood-free settings. We prove lower bounds on the predictive efficiency of the regions produced by CANVI and explore how the quality of a posterior approximation relates to the predictive efficiency of prediction regions based on that approximation. Finally, we demonstrate the accurate calibration and high predictive efficiency of CANVI on a suite of simulation-based inference benchmark tasks and an important scientific task: analyzing galaxy emission spectra.

Yash Patel

Many important computer vision tasks are naturally formulated to have a non-differentiable objective. Therefore, the standard, dominant training procedure of a neural network is not applicable since back-propagation requires the gradients of the objective with respect to the output of the model. Most deep learning methods side-step the problem sub-optimally by using a proxy loss for training, which was originally designed for another task and is not tailored to the specifics of the objective. The proxy loss functions may or may not align well with the original non-differentiable objective. An appropriate proxy has to be designed for a novel task, which may not be feasible for a non-specialist. This thesis makes four main contributions toward bridging the gap between the non-differentiable objective and the training loss function. Throughout the thesis, we refer to a loss function as a surrogate loss if it is a differentiable approximation of the non-differentiable objective. Note that we use the terms objective and evaluation metric interchangeably. The contributions of this thesis make the training of neural networks more scalable -- to new tasks in a nearly labor-free manner when the evaluation metric is decomposable, which will help researchers with novel tasks. For non-decomposable evaluation metrics, the differentiable components developed for the recall@k surrogate, such as sorting and counting, can also be used for creating new surrogates.

D. M. Anisuzzaman, Yash Patel, Behrouz Rostami et al.

Wound classification is an essential step of wound diagnosis. An efficient classifier can assist wound specialists in classifying wound types with less financial and time costs and help them decide an optimal treatment procedure. This study developed a deep neural network-based multi-modal classifier using wound images and their corresponding locations to categorize wound images into multiple classes, including diabetic, pressure, surgical, and venous ulcers. A body map is also developed to prepare the location data, which can help wound specialists tag wound locations more efficiently. Three datasets containing images and their corresponding location information are designed with the help of wound specialists. The multi-modal network is developed by concatenating the image-based and location-based classifier's outputs with some other modifications. The maximum accuracy on mixed-class classifications (containing background and normal skin) varies from 77.33% to 100% on different experiments. The maximum accuracy on wound-class classifications (containing only diabetic, pressure, surgical, and venous) varies from 72.95% to 98.08% on different experiments. The proposed multi-modal network also shows a significant improvement in results from the previous works of literature.

Yash Patel, Giorgos Tolias, Jiri Matas

This work focuses on learning deep visual representation models for retrieval by exploring the interplay between a new loss function, the batch size, and a new regularization approach. Direct optimization, by gradient descent, of an evaluation metric, is not possible when it is non-differentiable, which is the case for recall in retrieval. A differentiable surrogate loss for the recall is proposed in this work. Using an implementation that sidesteps the hardware constraints of the GPU memory, the method trains with a very large batch size, which is essential for metrics computed on the entire retrieval database. It is assisted by an efficient mixup regularization approach that operates on pairwise scalar similarities and virtually increases the batch size further. The suggested method achieves state-of-the-art performance in several image retrieval benchmarks when used for deep metric learning. For instance-level recognition, the method outperforms similar approaches that train using an approximation of average precision.

Yash Patel, Jiri Matas

This paper proposes a procedure to train a scene text recognition model using a robust learned surrogate of edit distance. The proposed method borrows from self-paced learning and filters out the training examples that are hard for the surrogate. The filtering is performed by judging the quality of the approximation, using a ramp function, enabling end-to-end training. Following the literature, the experiments are conducted in a post-tuning setup, where a trained scene text recognition model is tuned using the learned surrogate of edit distance. The efficacy is demonstrated by improvements on various challenging scene text datasets such as IIIT-5K, SVT, ICDAR, SVTP, and CUTE. The proposed method provides an average improvement of $11.2 \%$ on total edit distance and an error reduction of $9.5\%$ on accuracy.

Slobodan Djukanović, Yash Patel, Jiři Matas et al.

This paper addresses acoustic vehicle counting using one-channel audio. We predict the pass-by instants of vehicles from local minima of clipped vehicle-to-microphone distance. This distance is predicted from audio using a two-stage (coarse-fine) regression, with both stages realised via neural networks (NNs). Experiments show that the NN-based distance regression outperforms by far the previously proposed support vector regression. The $ 95\% $ confidence interval for the mean of vehicle counting error is within $[0.28\%, -0.55\%]$. Besides the minima-based counting, we propose a deep learning counting that operates on the predicted distance without detecting local minima. Although outperformed in accuracy by the former approach, deep counting has a significant advantage in that it does not depend on minima detection parameters. Results also show that removing low frequencies in features improves the counting performance.

D. M. Anisuzzaman, Yash Patel, Jeffrey Niezgoda et al.

We present an automated wound localizer from 2D wound and ulcer images by using deep neural network, as the first step towards building an automated and complete wound diagnostic system. The wound localizer has been developed by using YOLOv3 model, which is then turned into an iOS mobile application. The developed localizer can detect the wound and its surrounding tissues and isolate the localized wounded region from images, which would be very helpful for future processing such as wound segmentation and classification due to the removal of unnecessary regions from wound images. For Mobile App development with video processing, a lighter version of YOLOv3 named tiny-YOLOv3 has been used. The model is trained and tested on our own image dataset in collaboration with AZH Wound and Vascular Center, Milwaukee, Wisconsin. The YOLOv3 model is compared with SSD model, showing that YOLOv3 gives a mAP value of 93.9%, which is much better than the SSD model (86.4%). The robustness and reliability of these models are also tested on a publicly available dataset named Medetec and shows a very good performance as well.

Yash Patel, Tomas Hodan, Jiri Matas

This paper proposes a technique for training a neural network by minimizing a surrogate loss that approximates the target evaluation metric, which may be non-differentiable. The surrogate is learned via a deep embedding where the Euclidean distance between the prediction and the ground truth corresponds to the value of the evaluation metric. The effectiveness of the proposed technique is demonstrated in a post-tuning setup, where a trained model is tuned using the learned surrogate. Without a significant computational overhead and any bells and whistles, improvements are demonstrated on challenging and practical tasks of scene-text recognition and detection. In the recognition task, the model is tuned using a surrogate approximating the edit distance metric and achieves up to $39\%$ relative improvement in the total edit distance. In the detection task, the surrogate approximates the intersection over union metric for rotated bounding boxes and yields up to $4.25\%$ relative improvement in the $F_{1}$ score.

Yash Patel, Srikar Appalaraju, R. Manmatha

This paper proposes a new end-to-end trainable model for lossy image compression, which includes several novel components. The method incorporates 1) an adequate perceptual similarity metric; 2) saliency in the images; 3) a hierarchical auto-regressive model. This paper demonstrates that the popularly used evaluations metrics such as MS-SSIM and PSNR are inadequate for judging the performance of image compression techniques as they do not align with the human perception of similarity. Alternatively, a new metric is proposed, which is learned on perceptual similarity data specific to image compression. The proposed compression model incorporates the salient regions and optimizes on the proposed perceptual similarity metric. The model not only generates images which are visually better but also gives superior performance for subsequent computer vision tasks such as object detection and segmentation when compared to existing engineered or learned compression techniques.

Yash Patel, Srikar Appalaraju, R. Manmatha

Recently, there has been much interest in deep learning techniques to do image compression and there have been claims that several of these produce better results than engineered compression schemes (such as JPEG, JPEG2000 or BPG). A standard way of comparing image compression schemes today is to use perceptual similarity metrics such as PSNR or MS-SSIM (multi-scale structural similarity). This has led to some deep learning techniques which directly optimize for MS-SSIM by choosing it as a loss function. While this leads to a higher MS-SSIM for such techniques, we demonstrate using user studies that the resulting improvement may be misleading. Deep learning techniques for image compression with a higher MS-SSIM may actually be perceptually worse than engineered compression schemes with a lower MS-SSIM.

Yash Patel, Srikar Appalaraju, R. Manmatha

Several deep learned lossy compression techniques have been proposed in the recent literature. Most of these are optimized by using either MS-SSIM (multi-scale structural similarity) or MSE (mean squared error) as a loss function. Unfortunately, neither of these correlate well with human perception and this is clearly visible from the resulting compressed images. In several cases, the MS-SSIM for deep learned techniques is higher than say a conventional, non-deep learned codec such as JPEG-2000 or BPG. However, the images produced by these deep learned techniques are in many cases clearly worse to human eyes than those produced by JPEG-2000 or BPG. We propose the use of an alternative, deep perceptual metric, which has been shown to align better with human perceptual similarity. We then propose Deep Perceptual Compression (DPC) which makes use of an encoder-decoder based image compression model to jointly optimize on the deep perceptual metric and MS-SSIM. Via extensive human evaluations, we show that the proposed method generates visually better results than previous learning based compression methods and JPEG-2000, and is comparable to BPG. Furthermore, we demonstrate that for tasks like object-detection, images compressed with DPC give better accuracy.

Nibal Nayef, Yash Patel, Michal Busta et al.

With the growing cosmopolitan culture of modern cities, the need of robust Multi-Lingual scene Text (MLT) detection and recognition systems has never been more immense. With the goal to systematically benchmark and push the state-of-the-art forward, the proposed competition builds on top of the RRC-MLT-2017 with an additional end-to-end task, an additional language in the real images dataset, a large scale multi-lingual synthetic dataset to assist the training, and a baseline End-to-End recognition method. The real dataset consists of 20,000 images containing text from 10 languages. The challenge has 4 tasks covering various aspects of multi-lingual scene text: (a) text detection, (b) cropped word script classification, (c) joint text detection and script classification and (d) end-to-end detection and recognition. In total, the competition received 60 submissions from the research and industrial communities. This paper presents the dataset, the tasks and the findings of the presented RRC-MLT-2019 challenge.

Yash Patel, Lluis Gomez, Marçal Rusiñol et al.

Cross-modal retrieval methods have been significantly improved in last years with the use of deep neural networks and large-scale annotated datasets such as ImageNet and Places. However, collecting and annotating such datasets requires a tremendous amount of human effort and, besides, their annotations are usually limited to discrete sets of popular visual classes that may not be representative of the richer semantics found on large-scale cross-modal retrieval datasets. In this paper, we present a self-supervised cross-modal retrieval framework that leverages as training data the correlations between images and text on the entire set of Wikipedia articles. Our method consists in training a CNN to predict: (1) the semantic context of the article in which an image is more probable to appear as an illustration (global context), and (2) the semantic context of its caption (local context). Our experiments demonstrate that the proposed method is not only capable of learning discriminative visual representations for solving vision tasks like image classification and object detection, but that the learned representations are better for cross-modal retrieval when compared to supervised pre-training of the network on the ImageNet dataset.

Yash Patel, Lluis Gomez, Raul Gomez et al.

The immense success of deep learning based methods in computer vision heavily relies on large scale training datasets. These richly annotated datasets help the network learn discriminative visual features. Collecting and annotating such datasets requires a tremendous amount of human effort and annotations are limited to popular set of classes. As an alternative, learning visual features by designing auxiliary tasks which make use of freely available self-supervision has become increasingly popular in the computer vision community. In this paper, we put forward an idea to take advantage of multi-modal context to provide self-supervision for the training of computer vision algorithms. We show that adequate visual features can be learned efficiently by training a CNN to predict the semantic textual context in which a particular image is more probable to appear as an illustration. More specifically we use popular text embedding techniques to provide the self-supervision for the training of deep CNN. Our experiments demonstrate state-of-the-art performance in image classification, object detection, and multi-modal retrieval compared to recent self-supervised or naturally-supervised approaches.

Yash Patel, Kashyap Chitta, Bhavan Jasani

We address the problem of semi-supervised domain adaptation of classification algorithms through deep Q-learning. The core idea is to consider the predictions of a source domain network on target domain data as noisy labels, and learn a policy to sample from this data so as to maximize classification accuracy on a small annotated reward partition of the target domain. Our experiments show that learned sampling policies construct labeled sets that improve accuracies of visual classifiers over baselines.

Michal Bušta, Yash Patel, Jiri Matas

An end-to-end trainable (fully differentiable) method for multi-language scene text localization and recognition is proposed. The approach is based on a single fully convolutional network (FCN) with shared layers for both tasks. E2E-MLT is the first published multi-language OCR for scene text. While trained in multi-language setup, E2E-MLT demonstrates competitive performance when compared to other methods trained for English scene text alone. The experiments show that obtaining accurate multi-language multi-script annotations is a challenging problem.

Lluis Gomez, Yash Patel, Marçal Rusiñol et al.

End-to-end training from scratch of current deep architectures for new computer vision problems would require Imagenet-scale datasets, and this is not always possible. In this paper we present a method that is able to take advantage of freely available multi-modal content to train computer vision algorithms without human supervision. We put forward the idea of performing self-supervised learning of visual features by mining a large scale corpus of multi-modal (text and image) documents. We show that discriminative visual features can be learnt efficiently by training a CNN to predict the semantic context in which a particular image is more probable to appear as an illustration. For this we leverage the hidden semantic structures discovered in the text corpus with a well-known topic modeling technique. Our experiments demonstrate state of the art performance in image classification, object detection, and multi-modal retrieval compared to recent self-supervised or natural-supervised approaches.