arXiv subjects

Mertcan Cokbas

Publications and source records attributed to Mertcan Cokbas.

6 recordsLinked to original sources

arXiv2025

Improving Large-Scale Recommender Systems with Auxiliary Learning

Training large-scale recommendation models under a single global objective implicitly assumes homogeneity across user populations. However, real-world data are composites of heterogeneous cohorts with distinct conditional distributions. As models increase in scale and complexity and as more data is used for training, they become dominated by central distribution patterns, neglecting head and tail regions. This imbalance limits the model's learning ability and can result in inactive attention weights or dead neurons. In this paper, we reveal how the attention mechanism can play a key role in factorization machines for shared embedding selection, and propose to address this challenge by analyzing the substructures in the dataset and exposing those with strong distributional contrast through auxiliary learning. Unlike previous research, which heuristically applies weighted labels or multi-task heads to mitigate such biases, we leverage partially conflicting auxiliary labels to regularize the shared representation. This approach customizes the learning process of attention layers to preserve mutual information with minority cohorts while improving global performance. We evaluated proposed method on massive production datasets with billions of data points each for six SOTA models. Experiments show that the factorization machine is able to capture fine-grained user-ad interactions using the proposed method, achieving up to a 0.16% reduction in normalized entropy overall and delivering gains exceeding 0.30% on targeted minority cohorts.

cs.LG↗

arXiv2023

Complexity Evaluation of Parallel Execution of the RAPiD Deep-Learning Algorithm on Intel CPU

Knowing how many and where are people in various indoor spaces is critical for reducing HVAC energy waste, space management, spatial analytics and in emergency scenarios. While a range of technologies have been proposed to detect and track people in large indoor spaces, ceiling-mounted fisheye cameras have recently emerged as strong contenders. Currently, RAPiD is the SOTA algorithm for people detection in images captured by fisheye cameras. However, in large spaces several overhead fisheye cameras are needed to assure high accuracy of counting and thus multiple instances of RAPiD must be executed simultaneously. This report evaluates inference time when multiple instances of RAPiD run in parallel on an Ubuntu NUC PC with Intel I7 8559U CPU. We consider three mechanisms of CPU-resource allocation to handle multiple instances of RAPiD: 1) managed by Ubuntu, 2) managed by user via operating-system calls to assign logical cores, and 3) managed by user via PyTorch-library calls to limit the number of threads used by PyTorch. Each scenario was evaluated on 300 images. The experimental results show, that when one or two instances of RAPiD are executed in parallel all three approaches result in similar inference times of 1.8sec and 3.2sec, respectively. However, when three or more instances of RAPiD run in parallel, limiting the number of threads used by PyTorch results in the shortest inference times. On average, RAPiD completes inference of 2 images simultaneously in about 3sec, 4 images in 6sec and 8 images in less than 14sec. This is important for real-time system design. In HVAC-application scenarios, with a typical reaction time of 10-15min, a latency of 14sec is negligible so a single 8559U CPU can support 8 camera streams thus reducing the system cost. However, in emergency scenarios, when time is of essence, a single CPU may be needed for each camera to reduce the latency to 1.8sec.

cs.DC↗

arXiv2023

Estimating Distances Between People using a Single Overhead Fisheye Camera with Application to Social-Distancing Oversight

Unobtrusive monitoring of distances between people indoors is a useful tool in the fight against pandemics. A natural resource to accomplish this are surveillance cameras. Unlike previous distance estimation methods, we use a single, overhead, fisheye camera with wide area coverage and propose two approaches. One method leverages a geometric model of the fisheye lens, whereas the other method uses a neural network to predict the 3D-world distance from people-locations in a fisheye image. To evaluate our algorithms, we collected a first-of-its-kind dataset using single fisheye camera, that comprises a wide range of distances between people (1-58 ft) and will be made publicly available. The algorithms achieve 1-2 ft distance error and over 95% accuracy in detecting social-distance violations.

cs.CV↗

arXiv2022

Spatio-Visual Fusion-Based Person Re-Identification for Overhead Fisheye Images

Person re-identification (PRID) has been thoroughly researched in typical surveillance scenarios where various scenes are monitored by side-mounted, rectilinear-lens cameras. To date, few methods have been proposed for fisheye cameras mounted overhead and their performance is lacking. In order to close this performance gap, we propose a multi-feature framework for fisheye PRID where we combine deep-learning, color-based and location-based features by means of novel feature fusion. We evaluate the performance of our framework for various feature combinations on FRIDA, a public fisheye PRID dataset. The results demonstrate that our multi-feature approach outperforms recent appearance-based deep-learning methods by almost 18% points and location-based methods by almost 3% points in matching accuracy. We also demonstrate the potential application of the proposed PRID framework to people counting in large, crowded indoor spaces.

cs.CV↗

arXiv2022

FRIDA: Fisheye Re-Identification Dataset with Annotations

Person re-identification (PRID) from side-mounted rectilinear-lens cameras is a well-studied problem. On the other hand, PRID from overhead fisheye cameras is new and largely unstudied, primarily due to the lack of suitable image datasets. To fill this void, we introduce the "Fisheye Re-IDentification Dataset with Annotations" (FRIDA), with 240k+ bounding-box annotations of people, captured by 3 time-synchronized, ceiling-mounted fisheye cameras in a large indoor space. Due to a field-of-view overlap, PRID in this case differs from a typical PRID problem, which we discuss in depth. We also evaluate the performance of 10 state-of-the-art PRID algorithms on FRIDA. We show that for 6 CNN-based algorithms, training on FRIDA boosts the performance by up to 11.64% points in mAP compared to training on a common rectilinear-camera PRID dataset.

cs.CV↗

arXiv2020

Low-Resolution Overhead Thermal Tripwire for Occupancy Estimation

Smart buildings use occupancy sensing for various tasks ranging from energy-efficient HVAC and lighting to space-utilization analysis and emergency response. We propose a people counting system which uses a low-resolution thermal sensor. Unlike previous people-counting systems based on thermal sensors, we use an overhead tripwire configuration at entryways to detect and track transient entries or exits. We develop two distinct people counting algorithms for this configuration. To evaluate our algorithms, we have collected and labeled a low-resolution thermal video dataset using the proposed system. The dataset, the first of its kind, is public and available for download. We also propose new evaluation metrics that are more suitable for systems that are subject to drift and jitter.

cs.CV↗