Stephane Coulombe
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Stephane Coulombe has a total of 57 co-authors. Here are the top co-authors by number of shared publications:
Steven Pigeon
Department of Software and Information Technology Engineering, École de Technologie Supérieure, 1100 Notre-Dame Quest, Montreal, Quebec, H3C 1K3
8 shared publications- Carlos Vázquez
Laboratoire d'Innovation Ouverte (LIO), École de technologie supérieure, Montréal, Québec, Canada
6 shared publications E. Dubois
5 shared publicationsFrancois Caron
5 shared publications- Luc Trudeau
Department of Software and IT Engineering, École de technologie supérieure, Université du Québec, Montréal, Québec, Canada
5 shared publications - Christian Desrosiers
École de Technologie Supérieure
4 shared publications
Latest Publications
- Open Access
- 4 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 388-394
A Comprehensive Long-duration 8K Dataset to Benchmark Hardware Encoding for Live 360° Video Tiled Streaming
- 4 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 388-394
Open AccessA Comprehensive Long-duration 8K Dataset to Benchmark Hardware Encoding for Live 360° Video Tiled Streaming
- Open Access
- 10 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 122132-122149
Efficient Region-Wise Packing of Stereoscopic ERP Videos Based on Information Loss Minimization
Utilizing frame-compatible (FC) formats for packing stereoscopic videos often comes with challenges, as they require higher transmission bandwidth and larger memory buffers on the decoder compared to single-view videos. When it comes to stereoscopic 360° videos, as the primary content consumed by virtual reality (VR) applications, these requirements become even more challenging since they ask for ultra-high-resolution formats with high frame rates (e.g., 6K, 8K, or 12K at 100 frames per second). To address these challenges, sub-sampled versions of the left and right views are usually used to form the spatial FC format, leading to a loss of visual quality. In this paper, we propose an efficient region-wise packing method for equirectangular projection (ERP) videos with minimum information loss by exploiting the uneven sampling characteristic of ERP. Moreover, we propose a content-adaptive (CA) packing method for ERP videos, where the sizes of partitions, each with a particular horizontal downsampling factor, are adaptively determined based on spatial complexity. We then utilize a low-complexity frequency-domain approach to estimate the optimal partition sizes of the CA packing. We use these proposed methods to determine optimal packing of the stereoscopic ERP videos in the FC format. Experimental results, using the VVenC Versatile Video Coding (VVC) encoder, show that compared with the standard side-by-side (SbS) format, with uniform horizontal half-downsampling (UHHDS), the proposed CA packing method provides an average 13.84% and 12.02% Bjøntegaard-Delta bitrate (BD-BR) reduction for Random Access (RA) and Low Delay B (LDB) configurations, respectively, with an average encoding time comparable to SbS. In addition, when the performance is measured based on user attention probability, using the Laplacian Distribution model, the coding performance of our proposed packing methods outperforms the state-of-the-art packing method with significantly lower computational complexity.
- 10 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 122132-122149
Open AccessEfficient Region-Wise Packing of Stereoscopic ERP Videos Based on Information Loss Minimization
Utilizing frame-compatible (FC) formats for packing stereoscopic videos often comes with challenges, as they require higher transmission bandwidth and larger memory buffers on the decoder compared to single-view videos. When it comes to stereoscopic 360° videos, as the primary content consumed by virtual reality (VR) applications, these requirements become even more challenging since they ask for ultra-high-resolution formats with high frame rates (e.g., 6K, 8K, or 12K at 100 frames per second). To address these challenges, sub-sampled versions of the left and right views are usually used to form the spatial FC format, leading to a loss of visual quality. In this paper, we propose an efficient region-wise packing method for equirectangular projection (ERP) videos with minimum information loss by exploiting the uneven sampling characteristic of ERP. Moreover, we propose a content-adaptive (CA) packing method for ERP videos, where the sizes of partitions, each with a particular horizontal downsampling factor, are adaptively determined based on spatial complexity. We then utilize a low-complexity frequency-domain approach to estimate the optimal partition sizes of the CA packing. We use these proposed methods to determine optimal packing of the stereoscopic ERP videos in the FC format. Experimental results, using the VVenC Versatile Video Coding (VVC) encoder, show that compared with the standard side-by-side (SbS) format, with uniform horizontal half-downsampling (UHHDS), the proposed CA packing method provides an average 13.84% and 12.02% Bjøntegaard-Delta bitrate (BD-BR) reduction for Random Access (RA) and Low Delay B (LDB) configurations, respectively, with an average encoding time comparable to SbS. In addition, when the performance is measured based on user attention probability, using the Laplacian Distribution model, the coding performance of our proposed packing methods outperforms the state-of-the-art packing method with significantly lower computational complexity.
- Open Access
- 18 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 170632-170647
Robust Video List Decoding in Error-Prone Transmission Systems Using a Deep Learning Approach
This paper introduces a novel deep-learning assisted video list decoding method for error-prone video transmission systems. Unlike traditional list decoding techniques, our proposed system uses a Transformer-based no-reference image quality assessment method to select the highest-scoring reconstructed video candidate after reception. Three new components are defined and used in the Transformer-assisted image quality evaluation metric: neighborhood-based patch fidelity aggregation, discriminant color texture transformation and ranking-constrained penalty loss function.We have also created our own database of non-uniformly distorted images, similar to those that might result from transmission errors, in a High Efficiency Video Coding (HEVC) context. In our specific testing context, our improved Transformer-assisted method has a decision accuracy of 100% for intra-coded image, while, for errors occurring in an inter image, it is 96%. Notably, in the few cases where a wrong choice is made, the selected candidate’s quality remains similar to the intact frame. Code: https://github.com/Yujing0926/Robust-Video-List-Decoding-Using-a-Deep-Learning-Approach.
- 18 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 170632-170647
Open AccessRobust Video List Decoding in Error-Prone Transmission Systems Using a Deep Learning Approach
This paper introduces a novel deep-learning assisted video list decoding method for error-prone video transmission systems. Unlike traditional list decoding techniques, our proposed system uses a Transformer-based no-reference image quality assessment method to select the highest-scoring reconstructed video candidate after reception. Three new components are defined and used in the Transformer-assisted image quality evaluation metric: neighborhood-based patch fidelity aggregation, discriminant color texture transformation and ranking-constrained penalty loss function.We have also created our own database of non-uniformly distorted images, similar to those that might result from transmission errors, in a High Efficiency Video Coding (HEVC) context. In our specific testing context, our improved Transformer-assisted method has a decision accuracy of 100% for intra-coded image, while, for errors occurring in an inter image, it is 96%. Notably, in the few cases where a wrong choice is made, the selected candidate’s quality remains similar to the intact frame. Code: https://github.com/Yujing0926/Robust-Video-List-Decoding-Using-a-Deep-Learning-Approach.
- Open Access
- 26 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 140671-140680
No-Reference Video Quality Assessment Using Transformers and Attention Recurrent Networks
In recent years, numerous studies have investigated the development of methods for video quality assessment (VQA). These studies have predominantly focused on specific types of video degradation tailored to the application of interest. However, natural videos or recent videos generated by users (UGC) present complex distortions that are not easy to model. Consequently, most current VQA approaches struggle to achieve high performance when applied to these videos. In this paper, we propose a novel Transformerbased architecture that extracts spatial distortion features and spatio-temporal features from videos in two specialized branches. The spatial distortion branch leverages a transfer learning strategy where a standard ViT is pre-trained using a masked autoencoder (MAE) self-supervised learning task, and then fine-tuned to predict the distortion type of corrupted images from the CSIQ database. The features from this branch capture degradation at the level of individual frames. On the other hand, the second branch employs a 3D Shifted Windows Transformer (Swin-T) to extract spatio-temporal features across multiple frames. Once again, we use transfer learning to extract rich features by pre-training this 3D Swin-T model on a video dataset for human action recognition. Finally, a temporal memory block hinged on an attention recurrent neural networks is proposed to predict the final video quality score from the spatio-temporal sequence of features. We evaluate the performance of our method on two popular UGC databases, namely KoNViD-1k and LIVE-VQC. Results show it outperforms state-of-the-art models on the KoNViD-1k database, achieving a SROCC performance of 0.927 and a PLCC of 0.925, while also delivering highly competitive results on the LIVE-VQC database.
- 26 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 140671-140680
Open AccessNo-Reference Video Quality Assessment Using Transformers and Attention Recurrent Networks
In recent years, numerous studies have investigated the development of methods for video quality assessment (VQA). These studies have predominantly focused on specific types of video degradation tailored to the application of interest. However, natural videos or recent videos generated by users (UGC) present complex distortions that are not easy to model. Consequently, most current VQA approaches struggle to achieve high performance when applied to these videos. In this paper, we propose a novel Transformerbased architecture that extracts spatial distortion features and spatio-temporal features from videos in two specialized branches. The spatial distortion branch leverages a transfer learning strategy where a standard ViT is pre-trained using a masked autoencoder (MAE) self-supervised learning task, and then fine-tuned to predict the distortion type of corrupted images from the CSIQ database. The features from this branch capture degradation at the level of individual frames. On the other hand, the second branch employs a 3D Shifted Windows Transformer (Swin-T) to extract spatio-temporal features across multiple frames. Once again, we use transfer learning to extract rich features by pre-training this 3D Swin-T model on a video dataset for human action recognition. Finally, a temporal memory block hinged on an attention recurrent neural networks is proposed to predict the final video quality score from the spatio-temporal sequence of features. We evaluate the performance of our method on two popular UGC databases, namely KoNViD-1k and LIVE-VQC. Results show it outperforms state-of-the-art models on the KoNViD-1k database, achieving a SROCC performance of 0.927 and a PLCC of 0.925, while also delivering highly competitive results on the LIVE-VQC database.
- Open Access
- 14 April 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 41010-41022
No-Reference Video Quality Assessment Using Distortion Learning and Temporal Attention
The rapid growth of video consumption and multimedia applications has increased the interest of the academia and industry in building tools that can evaluate perceptual video quality. Since videos might be distorted when they are captured or transmitted, it is imperative to develop reliable methods for no-reference video quality assessment (NR-VQA). To date, most NR-VQA models in prior art have been proposed for assessing a specific category of distortion, such as authentic distortions or traditional distortions. Moreover, those developed for both authentic and traditional distortions video databases have so far led to poor performances. This resulted in the reluctance of service providers to adopt multiple NR-VQA approaches, as they prefer a single algorithm capable of accurately estimating video quality in all situations. Furthermore, many existing NR-VQA methods are computationally complex and therefore impractical for various real-life applications. In this paper, we propose a novel deep learning method for NR-VQA based on multi-task learning where the distortion of individual frames in a video and the overall quality of the video are predicted by a single neural network. This enables to train the network with a greater amount and variety of data, thereby improving its performance in testing. Additionally, our method leverages temporal attention to select the frames of a video sequence which contribute the most to its perceived quality. The proposed algorithm is evaluated on five publicly-available video quality assessment (VQA) databases containing traditional and authentic distortions. Results show that our method outperforms the state-of-the-art on traditional distortion databases such as LIVE VQA and CSIQ video, while also delivering competitive performance on databases containing authentic distortions such as KoNViD-1k, LIVE-Qualcomm and CVD2014.
- 14 April 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 41010-41022
Open AccessNo-Reference Video Quality Assessment Using Distortion Learning and Temporal Attention
The rapid growth of video consumption and multimedia applications has increased the interest of the academia and industry in building tools that can evaluate perceptual video quality. Since videos might be distorted when they are captured or transmitted, it is imperative to develop reliable methods for no-reference video quality assessment (NR-VQA). To date, most NR-VQA models in prior art have been proposed for assessing a specific category of distortion, such as authentic distortions or traditional distortions. Moreover, those developed for both authentic and traditional distortions video databases have so far led to poor performances. This resulted in the reluctance of service providers to adopt multiple NR-VQA approaches, as they prefer a single algorithm capable of accurately estimating video quality in all situations. Furthermore, many existing NR-VQA methods are computationally complex and therefore impractical for various real-life applications. In this paper, we propose a novel deep learning method for NR-VQA based on multi-task learning where the distortion of individual frames in a video and the overall quality of the video are predicted by a single neural network. This enables to train the network with a greater amount and variety of data, thereby improving its performance in testing. Additionally, our method leverages temporal attention to select the frames of a video sequence which contribute the most to its perceived quality. The proposed algorithm is evaluated on five publicly-available video quality assessment (VQA) databases containing traditional and authentic distortions. Results show that our method outperforms the state-of-the-art on traditional distortion databases such as LIVE VQA and CSIQ video, while also delivering competitive performance on databases containing authentic distortions such as KoNViD-1k, LIVE-Qualcomm and CVD2014.
- Open Access
- 1 March 2022
- Published by Elsevier in Signal Processing: Image Communication
- Vol. 102, 116597
Residual packet loss rate analysis of 2-D parity forward error correction
- 1 March 2022
- Published by Elsevier in Signal Processing: Image Communication
- Vol. 102, 116597
Open AccessResidual packet loss rate analysis of 2-D parity forward error correction
- Open Access
- 28 February 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 23931-23947
CRC-Based Correction of Multiple Errors Using an Optimized Lookup Table
In this paper, we propose a new approach to perform multiple error correction in wireless communications over error-prone networks. It is based on the cyclic redundancy check syndrome, using an optimized lookup table that avoids performing arithmetic operations. This method is able to achieve the same correction performance as the state-of-the-art approaches while significantly reducing the computational complexity. The table is designed to allow multiple bit error correction simply by navigating within it. Its size is constant when considering more than two errors, which represents a tremendous advantage over earlier lookup table-based approaches. Simulation results of a C implementation performed on a Raspberry Pi 4 show that the proposed method is able to process single and double error corrections of large payloads in 100ns and 642 $\mu \text{s}$ , respectively, while it would take 300 $\mu \text{s}$ and 1.5s, respectively, with the state-of-the-art CRC multiple error correction technique. This represents a speedup of nearly $3000\pmb {\times }$ for single error and $2300\pmb {\times }$ for double error correction, respectively. Compared to table-based approaches, the proposed method offers a speedup of nearly $1200\pmb {\times }$ for single error and $2300\pmb {\times }$ for double error correction under the same conditions. We also show that when multiple candidate error patterns are present, numerous errors can be corrected by adding a checksum cross-validation step.
- 28 February 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 23931-23947
Open AccessCRC-Based Correction of Multiple Errors Using an Optimized Lookup Table
In this paper, we propose a new approach to perform multiple error correction in wireless communications over error-prone networks. It is based on the cyclic redundancy check syndrome, using an optimized lookup table that avoids performing arithmetic operations. This method is able to achieve the same correction performance as the state-of-the-art approaches while significantly reducing the computational complexity. The table is designed to allow multiple bit error correction simply by navigating within it. Its size is constant when considering more than two errors, which represents a tremendous advantage over earlier lookup table-based approaches. Simulation results of a C implementation performed on a Raspberry Pi 4 show that the proposed method is able to process single and double error corrections of large payloads in 100ns and 642 $\mu \text{s}$ , respectively, while it would take 300 $\mu \text{s}$ and 1.5s, respectively, with the state-of-the-art CRC multiple error correction technique. This represents a speedup of nearly $3000\pmb {\times }$ for single error and $2300\pmb {\times }$ for double error correction, respectively. Compared to table-based approaches, the proposed method offers a speedup of nearly $1200\pmb {\times }$ for single error and $2300\pmb {\times }$ for double error correction under the same conditions. We also show that when multiple candidate error patterns are present, numerous errors can be corrected by adding a checksum cross-validation step.
- 1 November 2021
- Published by Elsevier in Signal Processing: Image Communication
- Vol. 99
Enhanced CRC-based correction of multiple errors with candidate validation
- 1 November 2021
- Published by Elsevier in Signal Processing: Image Communication
- Vol. 99
Enhanced CRC-based correction of multiple errors with candidate validation
- Open Access
- 1 June 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 102357-102372
Table-Free Multiple Bit-Error Correction Using the CRC Syndrome
In this paper, we propose a novel method for correcting multiple errors in data packets, using the Cyclic Redundancy Check (CRC) syndrome present in low layers of protocol stacks. The proposed method generates the whole list of error patterns, leading to a received syndrome containing up to a given maximum number of errors. Our approach is table-free, is computationally efficient, and can instantly correct erroneous packets when the output list contains a single element. A performance study is conducted, and shows that the proposed approach outperforms existing ones in Bluetooth Low Energy (BLE) as it can correct all single- and double-error patterns as well as most triple-error cases when considering small payloads used in Internet of Things (IoT) applications.
- 1 June 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 102357-102372
Open AccessTable-Free Multiple Bit-Error Correction Using the CRC Syndrome
In this paper, we propose a novel method for correcting multiple errors in data packets, using the Cyclic Redundancy Check (CRC) syndrome present in low layers of protocol stacks. The proposed method generates the whole list of error patterns, leading to a received syndrome containing up to a given maximum number of errors. Our approach is table-free, is computationally efficient, and can instantly correct erroneous packets when the output list contains a single element. A performance study is conducted, and shows that the proposed approach outperforms existing ones in Bluetooth Low Energy (BLE) as it can correct all single- and double-error patterns as well as most triple-error cases when considering small payloads used in Internet of Things (IoT) applications.
- Open Access
- 8 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 70234-70244
Power Modeling for Video Streaming Applications on Mobile Devices
In this paper, we derive an accurate power model for video streaming which we condense to the essential components contributing the most to the overall power consumption. As a use case, we choose mobile devices on the receiver side performing video streaming in broadcasting or end-to-end scenarios. In modeling, we consider the complete video streaming toolchain, which mainly consists of data acquisition, video processing, display, and audio handling. We compose an overall power model with the help of models from the literature and propose a dedicated feature selection approach to reveal the most important factors related to power consumption. The resulting models achieve mean estimation errors below 7.61%. Results from feature selection indicate that the display brightness, the bitrate, and the frame rate have the highest impact on the power consumption.
- 8 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 70234-70244
Open AccessPower Modeling for Video Streaming Applications on Mobile Devices
In this paper, we derive an accurate power model for video streaming which we condense to the essential components contributing the most to the overall power consumption. As a use case, we choose mobile devices on the receiver side performing video streaming in broadcasting or end-to-end scenarios. In modeling, we consider the complete video streaming toolchain, which mainly consists of data acquisition, video processing, display, and audio handling. We compose an overall power model with the help of models from the literature and propose a dedicated feature selection approach to reveal the most important factors related to power consumption. The resulting models achieve mean estimation errors below 7.61%. Results from feature selection indicate that the display brightness, the bitrate, and the frame rate have the highest impact on the power consumption.