Ankit Kumar

Xin Li, Yeying Jin, Suhang Yao et al.

This paper presents an overview of the NTIRE 2026 Second Challenge on Day and Night Raindrop Removal for Dual-Focused Images. Building upon the success of the first edition, this challenge attracted a wide range of impressive solutions, all developed and evaluated on our real-world Raindrop Clarity dataset~\cite{jin2024raindrop}. For this edition, we adjust the dataset with 14,139 images for training, 407 images for validation, and 593 images for testing. The primary goal of this challenge is to establish a strong and practical benchmark for the removal of raindrops under various illumination and focus conditions. In total, 168 teams have registered for the competition, and 17 teams submitted valid final solutions and fact sheets for the testing phase. The submitted methods achieved strong performance on the Raindrop Clarity dataset, demonstrating the growing progress in this challenging task.

Xinyu Zhang, Ziyi Kou, Chuan Qin et al.

Understanding hand-object interaction (HOI) is fundamental to computer vision, robotics, and AR/VR. However, conventional hand videos often lack essential physical information such as contact forces and motion signals, and are prone to frequent occlusions. To address the challenges, we present Glove2Hand, a framework that translates multi-modal sensing glove HOI videos into photorealistic bare hands, while faithfully preserving the underlying physical interaction dynamics. We introduce a novel 3D Gaussian hand model that ensures temporal rendering consistency. The rendered hand is seamlessly integrated into the scene using a diffusion-based hand restorer, which effectively handles complex hand-object interactions and non-rigid deformations. Leveraging Glove2Hand, we create HandSense, the first multi-modal HOI dataset featuring glove-to-hand videos with synchronized tactile and IMU signals. We demonstrate that HandSense significantly enhances downstream bare-hand applications, including video-based contact estimation and hand tracking under severe occlusion.

Ziyi Kou, Ankit Kumar, Mia Huang et al.

We present AVI-HT, an adaptive visual-IMU fusion approach for tracking 3D hand poses by jointly modeling the egocentric image with on-glove 6-DoF IMU signals. AVI-HT achieves significantly improved accuracy and availability, particularly in hand-object interaction (HOI) scenarios involving heavy visual occlusion. Two complementary ingredients underpin its success: (1) synchronized multi-modal training data pairing on-body vision-IMU sensor streams with ground-truth 3D hand poses from a motion-capture system, and (2) a cross-sensor deep attention mechanism that adaptively modulates the trust assigned to the vision and individual IMU sensors. To evaluate AVI-HT in real-world settings, we conduct extensive experiments on our DexGloveHOI dataset that consists of 100K+ pairwise vision-IMU samples with synchronized 3D annotated poses, in which users manipulate a variety of objects during daily tasks. We compare against multiple single- and multi-modal tracking approaches under two hand models (UmeTrack, MANO). The results show that AVI-HT reduces mean keypoint error by 16.1% and its wrist-aligned variant by 24.2% over the baselines. Ablation studies further reveal the per-finger contribution of IMU sensors across activity types, and the model's sensitivity to IMU noise and temporal misalignment in vision-IMU fusion.

Ankit Kumar, Priya Shukla, Vandana Kushwaha et al.

6D object pose estimation has been a research topic in the field of computer vision and robotics. Many modern world applications like robot grasping, manipulation, autonomous navigation etc, require the correct pose of objects present in a scene to perform their specific task. It becomes even harder when the objects are placed in a cluttered scene and the level of occlusion is high. Prior works have tried to overcome this problem but could not achieve accuracy that can be considered reliable in real-world applications. In this paper, we present an architecture that, unlike prior work, is context-aware. It utilizes the context information available to us about the objects. Our proposed architecture treats the objects separately according to their types i.e; symmetric and non-symmetric. A deeper estimator and refiner network pair is used for non-symmetric objects as compared to symmetric due to their intrinsic differences. Our experiments show an enhancement in the accuracy of about 3.2% over the LineMOD dataset, which is considered a benchmark for pose estimation in the occluded and cluttered scenes, against the prior state-of-the-art DenseFusion. Our results also show that the inference time we got is sufficient for real-time usage.

Yashwardhan Chaudhuri, Ankit Kumar, Arun Balaji Buduru et al.

Crowd counting is gaining societal relevance, particularly in domains of Urban Planning, Crowd Management, and Public Safety. This paper introduces Fourier-guided attention (FGA), a novel attention mechanism for crowd count estimation designed to address the inefficient full-scale global pattern capture in existing works on convolution-based attention networks. FGA efficiently captures multi-scale information, including full-scale global patterns, by utilizing Fast-Fourier Transformations (FFT) along with spatial attention for global features and convolutions with channel-wise attention for semi-global and local features. The architecture of FGA involves a dual-path approach: (1) a path for processing full-scale global features through FFT, allowing for efficient extraction of information in the frequency domain, and (2) a path for processing remaining feature maps for semi-global and local features using traditional convolutions and channel-wise attention. This dual-path architecture enables FGA to seamlessly integrate frequency and spatial information, enhancing its ability to capture diverse crowd patterns. We apply FGA in the last layers of two popular crowd-counting works, CSRNet and CANNet, to evaluate the module's performance on benchmark datasets such as ShanghaiTech-A, ShanghaiTech-B, UCF-CC-50, and JHU++ crowd. The experiments demonstrate a notable improvement across all datasets based on Mean-Squared-Error (MSE) and Mean-Absolute-Error (MAE) metrics, showing comparable performance to recent state-of-the-art methods. Additionally, we illustrate the interpretability using qualitative analysis, leveraging Grad-CAM heatmaps, to show the effectiveness of FGA in capturing crowd patterns.

Pravneet Kaur, Gautam Siddharth Kashyap, Ankit Kumar et al.

This groundbreaking study explores the expanse of Large Language Models (LLMs), such as Generative Pre-Trained Transformer (GPT) and Bidirectional Encoder Representations from Transformers (BERT) across varied domains ranging from technology, finance, healthcare to education. Despite their established prowess in Natural Language Processing (NLP), these LLMs have not been systematically examined for their impact on domains such as fitness, and holistic well-being, urban planning, climate modelling as well as disaster management. This review paper, in addition to furnishing a comprehensive analysis of the vast expanse and extent of LLMs' utility in diverse domains, recognizes the research gaps and realms where the potential of LLMs is yet to be harnessed. This study uncovers innovative ways in which LLMs can leave a mark in the fields like fitness and wellbeing, urban planning, climate modelling and disaster response which could inspire future researches and applications in the said avenues.

Yashwardhan Chaudhuri, Ankit Kumar, Orchid Chetia Phukan et al.

Crowd counting finds direct applications in real-world situations, making computational efficiency and performance crucial. However, most of the previous methods rely on a heavy backbone and a complex downstream architecture that restricts the deployment. To address this challenge and enhance the versatility of crowd-counting models, we introduce two lightweight models. These models maintain the same downstream architecture while incorporating two distinct backbones: MobileNet and MobileViT. We leverage Adjacent Feature Fusion to extract diverse scale features from a Pre-Trained Model (PTM) and subsequently combine these features seamlessly. This approach empowers our models to achieve improved performance while maintaining a compact and efficient design. With the comparison of our proposed models with previously available state-of-the-art (SOTA) methods on ShanghaiTech-A ShanghaiTech-B and UCF-CC-50 dataset, it achieves comparable results while being the most computationally efficient model. Finally, we present a comparative study, an extensive ablation study, along with pruning to show the effectiveness of our models.

Zheng Chen, Kai Liu, Jue Gong et al.

This paper presents the NTIRE 2025 image super-resolution ($\times$4) challenge, one of the associated competitions of the 10th NTIRE Workshop at CVPR 2025. The challenge aims to recover high-resolution (HR) images from low-resolution (LR) counterparts generated through bicubic downsampling with a $\times$4 scaling factor. The objective is to develop effective network designs or solutions that achieve state-of-the-art SR performance. To reflect the dual objectives of image SR research, the challenge includes two sub-tracks: (1) a restoration track, emphasizes pixel-wise accuracy and ranks submissions based on PSNR; (2) a perceptual track, focuses on visual realism and ranks results by a perceptual score. A total of 286 participants registered for the competition, with 25 teams submitting valid entries. This report summarizes the challenge design, datasets, evaluation protocol, the main results, and methods of each team. The challenge serves as a benchmark to advance the state of the art and foster progress in image SR.

Kartikay Bagla, Ankit Kumar, Shivam Gupta et al.

In the last few years, the ML community has created a number of new NLP models based on transformer architecture. These models have shown great performance for various NLP tasks on benchmark datasets, often surpassing SOTA results. Buoyed with this success, one often finds industry practitioners actively experimenting with fine-tuning these models to build NLP applications for industry use cases. However, for most datasets that are used by practitioners to build industrial NLP applications, it is hard to guarantee the presence of any noise in the data. While most transformer based NLP models have performed exceedingly well in transferring the learnings from one dataset to another, it remains unclear how these models perform when fine-tuned on noisy text. We address the open question by Kumar et al. (2020) to explore the sensitivity of popular transformer based NLP models to noise in the text data. We continue working with the noise as defined by them -- spelling mistakes & typos (which are the most commonly occurring noise). We show (via experimental results) that these models perform badly on most common NLP tasks namely text classification, textual similarity, NER, question answering, text summarization on benchmark datasets. We further show that as the noise in data increases, the performance degrades. Our findings suggest that one must be vary of the presence of noise in their datasets while fine-tuning popular transformer based NLP models.

Sukhdeep S. Sodhi, Ellie Ka-In Chio, Ambarish Jash et al.

As more and more online search queries come from voice, automatic speech recognition becomes a key component to deliver relevant search results. Errors introduced by automatic speech recognition (ASR) lead to irrelevant search results returned to the user, thus causing user dissatisfaction. In this paper, we introduce an approach, Mondegreen, to correct voice queries in text space without depending on audio signals, which may not always be available due to system constraints or privacy or bandwidth (for example, some ASR systems run on-device) considerations. We focus on voice queries transcribed via several proprietary commercial ASR systems. These queries come from users making internet, or online service search queries. We first present an analysis showing how different the language distribution coming from user voice queries is from that in traditional text corpora used to train off-the-shelf ASR systems. We then demonstrate that Mondegreen can achieve significant improvements in increased user interaction by correcting user voice queries in one of the largest search systems in Google. Finally, we see Mondegreen as complementing existing highly-optimized production ASR systems, which may not be frequently retrained and thus lag behind due to vocabulary drifts.

Andrés L. Suárez-Cetrulo, Ankit Kumar, Luis Miralles-Pechuán

The investment of time and resources for better strategies and methodologies to tackle a potential pandemic is key to deal with potential outbreaks of new variants or other viruses in the future. In this work, we recreated the scene of a year ago, 2020, when the pandemic erupted across the world for the fifty countries with more COVID-19 cases reported. We performed some experiments in which we compare state-of-the-art machine learning algorithms, such as LSTM, against online incremental machine learning algorithms to adapt them to the daily changes in the spread of the disease and predict future COVID-19 cases. To compare the methods, we performed three experiments: In the first one, we trained the models using only data from the country we predicted. In the second one, we use data from all fifty countries to train and predict each of them. In the first and second experiment, we used a static hold-out approach for all methods. In the third experiment, we trained the incremental methods sequentially, using a prequential evaluation. This scheme is not suitable for most state-of-the-art machine learning algorithms because they need to be retrained from scratch for every batch of predictions, causing a computational burden. Results show that incremental methods are a promising approach to adapt to changes of the disease over time; they are always up to date with the last state of the data distribution, and they have a significantly lower computational cost than other techniques such as LSTMs.

Piyush Makhija, Ankit Kumar, Anuj Gupta

We present hinglishNorm -- a human annotated corpus of Hindi-English code-mixed sentences for text normalization task. Each sentence in the corpus is aligned to its corresponding human annotated normalized form. To the best of our knowledge, there is no corpus of Hindi-English code-mixed sentences for text normalization task that is publicly available. Our work is the first attempt in this direction. The corpus contains 13494 parallel segments. Further, we present baseline normalization results on this corpus. We obtain a Word Error Rate (WER) of 15.55, BiLingual Evaluation Understudy (BLEU) score of 71.2, and Metric for Evaluation of Translation with Explicit ORdering (METEOR) score of 0.50.

Tao Wu, Ellie Ka-In Chio, Heng-Tze Cheng et al.

Many recent advances in neural information retrieval models, which predict top-K items given a query, learn directly from a large training set of (query, item) pairs. However, they are often insufficient when there are many previously unseen (query, item) combinations, often referred to as the cold start problem. Furthermore, the search system can be biased towards items that are frequently shown to a query previously, also known as the 'rich get richer' (a.k.a. feedback loop) problem. In light of these problems, we observed that most online content platforms have both a search and a recommender system that, while having heterogeneous input spaces, can be connected through their common output item space and a shared semantic representation. In this paper, we propose a new Zero-Shot Heterogeneous Transfer Learning framework that transfers learned knowledge from the recommender system component to improve the search component of a content platform. First, it learns representations of items and their natural-language features by predicting (item, item) correlation graphs derived from the recommender system as an auxiliary task. Then, the learned representations are transferred to solve the target search retrieval task, performing query-to-item prediction without having seen any (query, item) pairs in training. We conduct online and offline experiments on one of the world's largest search and recommender systems from Google, and present the results and lessons learned. We demonstrate that the proposed approach can achieve high performance on offline search retrieval tasks, and more importantly, achieved significant improvements on relevance and user interactions over the highly-optimized production system in online experiments.

Ali Hadian, Ankit Kumar, Thomas Heinis

Spatial indexes are crucial for the analysis of the increasing amounts of spatial data, for example generated through IoT applications. The plethora of indexes that has been developed in recent decades has primarily been optimised for disk. With increasing amounts of memory even on commodity machines, however, moving them to main memory is an option. Doing so opens up the opportunity to use additional optimizations that are only amenable to main memory. In this paper we thus explore the opportunity to use light-weight machine learning models to accelerate queries on spatial indexes. We do so by exploring the potential of using interpolation and similar techniques on the R-tree, arguably the most broadly used spatial index. As we show in our experimental analysis, the query execution time can be reduced by up to 60% while simultaneously shrinking the index's memory footprint by over 90%

Ankit Kumar, Piyush Makhija, Anuj Gupta

Owing to the phenomenal success of BERT on various NLP tasks and benchmark datasets, industry practitioners are actively experimenting with fine-tuning BERT to build NLP applications for solving industry use cases. For most datasets that are used by practitioners to build industrial NLP applications, it is hard to guarantee absence of any noise in the data. While BERT has performed exceedingly well for transferring the learnings from one use case to another, it remains unclear how BERT performs when fine-tuned on noisy text. In this work, we explore the sensitivity of BERT to noise in the data. We work with most commonly occurring noise (spelling mistakes, typos) and show that this results in significant degradation in the performance of BERT. We present experimental results to show that BERT's performance on fundamental NLP tasks like sentiment analysis and textual similarity drops significantly in the presence of (simulated) noise on benchmark datasets viz. IMDB Movie Review, STS-B, SST-2. Further, we identify shortcomings in the existing BERT pipeline that are responsible for this drop in performance. Our findings suggest that practitioners need to be vary of presence of noise in their datasets while fine-tuning BERT to solve industry use cases.

Shweta Yadav, Ankit Kumar, Asif Ekbal et al.

Knowledge about protein-protein interactions is essential in understanding the biological processes such as metabolic pathways, DNA replication, and transcription etc. However, a majority of the existing Protein-Protein Interaction (PPI) systems are dependent primarily on the scientific literature, which is yet not accessible as a structured database. Thus, efficient information extraction systems are required for identifying PPI information from the large collection of biomedical texts. Most of the existing systems model the PPI extraction task as a classification problem and are tailored to the handcrafted feature set including domain dependent features. In this paper, we present a novel method based on deep bidirectional long short-term memory (B-LSTM) technique that exploits word sequences and dependency path related information to identify PPI information from text. This model leverages joint modeling of proteins and relations in a single unified framework, which we name as Shortest Dependency Path B-LSTM (sdpLSTM) model. We perform experiments on two popular benchmark PPI datasets, namely AiMed & BioInfer. The evaluation shows the F1-score values of 86.45% and 77.35% on AiMed and BioInfer, respectively. Comparisons with the existing systems show that our proposed approach attains state-of-the-art performance.

Sourish Ghosh, Aaditya Sanjay Boob, Nishant Nikhil et al.

A college student's life can be primarily categorized into domains such as education, health, social and other activities which may include daily chores and travelling time. Time management is crucial for every student. A self realisation of one's daily time expenditure in various domains is therefore essential to maximize one's effective output. This paper presents how a mobile application using Fuzzy Logic and Global Positioning System (GPS) analyzes a student's lifestyle and provides recommendations and suggestions based on the results.

Malladihalli S Bhuvan, Ankit Kumar, Adil Zafar et al.

Hospital readmissions are expensive and reflect the inadequacies in healthcare system. In the United States alone, treatment of readmitted diabetic patients exceeds 250 million dollars per year. Early identification of patients facing a high risk of readmission can enable healthcare providers to to conduct additional investigations and possibly prevent future readmissions. This not only improves the quality of care but also reduces the medical expenses on readmission. Machine learning methods have been leveraged on public health data to build a system for identifying diabetic patients facing a high risk of future readmission. Number of inpatient visits, discharge disposition and admission type were identified as strong predictors of readmission. Further, it was found that the number of laboratory tests and discharge disposition together predict whether the patient will be readmitted shortly after being discharged from the hospital (i.e. <30 days) or after a longer period of time (i.e. >30 days). These insights can help healthcare providers to improve inpatient diabetic care. Finally, the cost analysis suggests that \$252.76 million can be saved across 98,053 diabetic patient encounters by incorporating the proposed cost sensitive analysis model.

Joy Bose, Amit Singhai, Anish Patankar et al.

With a number of cheap commercial dry EEG kits available today, it is possible to look at user attention driven scenarios for interaction with the web browser. Using EEG to determine the user's attention level is preferable to using methods such as gaze tracking or time spent on the webpage. In this paper we use the attention level in three different ways. First, as a control mechanism, to control user interface elements such as menus or buttons. Second, to make the web browser responsive to the current attention level. Third, as a means for the web developer to control the user experience based on the level of attention paid by the user, thus creating attention sensitive websites. We present implementation details for each of these, using the NeuroSky MindWave sensor. We also explore issues in the system, and possibility of an EEG based web standard.

Ankit Kumar, Ozan Irsoy, Peter Ondruska et al.

Most tasks in natural language processing can be cast into question answering (QA) problems over language input. We introduce the dynamic memory network (DMN), a neural network architecture which processes input sequences and questions, forms episodic memories, and generates relevant answers. Questions trigger an iterative attention process which allows the model to condition its attention on the inputs and the result of previous iterations. These results are then reasoned over in a hierarchical recurrent sequence model to generate answers. The DMN can be trained end-to-end and obtains state-of-the-art results on several types of tasks and datasets: question answering (Facebook's bAbI dataset), text classification for sentiment analysis (Stanford Sentiment Treebank) and sequence modeling for part-of-speech tagging (WSJ-PTB). The training for these different tasks relies exclusively on trained word vector representations and input-question-answer triplets.

Ankit Kumar, Tushar Patnaik, Vivek Kr Verma

India is a multilingual multi-script country. In every state of India there are two languages one is state local language and the other is English. For example in Andhra Pradesh, a state in India, the document may contain text words in English and Telugu script. For Optical Character Recognition (OCR) of such a bilingual document, it is necessary to identify the script before feeding the text words to the OCRs of individual scripts. In this paper, we are introducing a simple and efficient technique of script identification for Kannada, English and Hindi text words of a printed document. The proposed approach is based on the horizontal and vertical projection profile for the discrimination of the three scripts. The feature extraction is done based on the horizontal projection profile of each text words. We analysed 700 different words of Kannada, English and Hindi in order to extract the discrimination features and for the development of knowledge base. We use the horizontal projection profile of each text word and based on the horizontal projection profile we extract the appropriate features. The proposed system is tested on 100 different document images containing more than 1000 text words of each script and a classification rate of 98.25%, 99.25% and 98.87% is achieved for Kannada, English and Hindi respectively.