Punnarai Siricharoen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Punnarai Siricharoen has a total of 33 co-authors. Here are the top co-authors by number of shared publications:
- Krittiya Phitchakian
INFRA PLUS, Infra Plus Co., Ltd., Bangkok, Thailand
2 shared publications Thidarat Bunsri
2 shared publicationsWarisa Yomsatieankul
2 shared publications- Chanakarn Suebthawinkul
Department of Obstetrics and Gynecology, Faculty of Medicine, Chulalongkorn University, Bangkok, Thailand
1 shared publication - Kasem Choocharukul1 shared publication
- Kerkrith NaMuengklai
INFRA PLUS, Infra Plus Co., Ltd., Bangkok, Thailand
1 shared publication
Latest Publications
- Open Access
- 5 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 85709-85723
YOWOFormer: Bridging Video Transformers and CNN Detectors for One-Stage Spatio-Temporal Action Detection
The YOWO family provides one-stage dual-stream detectors for spatio-temporal action detection (STAD) by decoupling spatial (2D CNN) and temporal (3D CNN) backbones, enabling modality-specific pre-training and independent optimization. All existing variants, however, rely on 3D CNNs whose temporal receptive field is limited. Video transformers such as VideoMAE capture richer temporal representations, yet their token outputs are not directly compatible with CNN-based detectors. We present YOWOFormer, which replaces the 3D CNN stream with VideoMAE and introduces two bridging modules. Specifically, a Spatial Query Adapter (SQA) converts transformer tokens into spatial feature maps via learnable queries and cross-attention, and a Bidirectional Spatio-Temporal Fusion (BSTF) enables spatial and temporal features to mutually reinforce through bidirectional cross-attention.We further conduct systematic comparisons across four adapter designs, three fusion strategies, and the scaling of backbone capacity and clip length. On UCF101-24, YOWOFormer-L achieves 93.35% frame-level mAP, a new state of the art surpassing the previous best of 88.64%, and YOWOFormer-B attains 91.47% at 26.81 frames per second (FPS). On AVA v2.2, YOWOFormer-B and YOWOFormer-L improve by 2.34 and 5.39 percentage points over prior dual-stream 3D-CNN-based methods, respectively. Ablation studies demonstrate that temporal representation is the primary bottleneck in dual-stream STAD and that scaling video transformers yields more consistent gains than scaling 3D CNNs.
- 5 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 85709-85723
Open AccessYOWOFormer: Bridging Video Transformers and CNN Detectors for One-Stage Spatio-Temporal Action Detection
The YOWO family provides one-stage dual-stream detectors for spatio-temporal action detection (STAD) by decoupling spatial (2D CNN) and temporal (3D CNN) backbones, enabling modality-specific pre-training and independent optimization. All existing variants, however, rely on 3D CNNs whose temporal receptive field is limited. Video transformers such as VideoMAE capture richer temporal representations, yet their token outputs are not directly compatible with CNN-based detectors. We present YOWOFormer, which replaces the 3D CNN stream with VideoMAE and introduces two bridging modules. Specifically, a Spatial Query Adapter (SQA) converts transformer tokens into spatial feature maps via learnable queries and cross-attention, and a Bidirectional Spatio-Temporal Fusion (BSTF) enables spatial and temporal features to mutually reinforce through bidirectional cross-attention.We further conduct systematic comparisons across four adapter designs, three fusion strategies, and the scaling of backbone capacity and clip length. On UCF101-24, YOWOFormer-L achieves 93.35% frame-level mAP, a new state of the art surpassing the previous best of 88.64%, and YOWOFormer-B attains 91.47% at 26.81 frames per second (FPS). On AVA v2.2, YOWOFormer-B and YOWOFormer-L improve by 2.34 and 5.39 percentage points over prior dual-stream 3D-CNN-based methods, respectively. Ablation studies demonstrate that temporal representation is the primary bottleneck in dual-stream STAD and that scaling video transformers yields more consistent gains than scaling 3D CNNs.
- Open Access
- 29 December 2025
- Published by Atlantis Press SARL
- p. 188-198
Enhancing Highway Traffic Volume Surveys in Thailand with Image Processing and Machine Learning
- 29 December 2025
- Published by Atlantis Press SARL
- p. 188-198
Open AccessEnhancing Highway Traffic Volume Surveys in Thailand with Image Processing and Machine Learning
- Open Access
- 28 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 51-63
Diffusion-Based Synthetic Augmentation with Quality-Aware Reference Selection and mAP-Guided Allocation for Enhancing Vehicle Detection Robustness
- 28 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 51-63
Open AccessDiffusion-Based Synthetic Augmentation with Quality-Aware Reference Selection and mAP-Guided Allocation for Enhancing Vehicle Detection Robustness
- Open Access
- 23 March 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-16
A paired multi-scale attention network for liver tumor segmentation in 99mTc-MAA SPECT/CT imaging
Automatic liver tumor segmentation using Single-Photon Emission Computed Tomography/Computed Tomography (SPECT/CT) provides detailed insights that facilitate accurate tumor targeting and effective treatment planning. However, challenges such as spill-out can distort tumor size, which complicates accurate segmentation. This study introduces a Paired Multi-Scale Attention Network (P-MANet) using Top-Hat Cross-Features to address these challenges by enhancing the identification of true positives while minimizing false positives in liver tumor segmentation. P-MANet employs a dual-branch architecture. The first branch utilizes a Multi-Scale Attention Network (MA-Net) to process SPECT/CT datasets, whereas the second branch applies the White Top-Hat Transform to extract features that are then integrated with those from the first branch. This innovative approach effectively mitigates errors stemming from spectral light variation, leading to improved accuracy in tumor delineation. The model was trained on a dataset comprising 43 cases of 99mTc-MAA SPECT/CT. P-MANet achieved a Dice Similarity Coefficient (DSC) of 67.93% and 66.56% for normal and abnormal spectral light distributions, respectively, and obtained a DSC of 67.00%, on average, which outperformed other models, including results from a previous study that was tested on the same dataset.
- 23 March 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-16
Open AccessA paired multi-scale attention network for liver tumor segmentation in 99mTc-MAA SPECT/CT imaging
Automatic liver tumor segmentation using Single-Photon Emission Computed Tomography/Computed Tomography (SPECT/CT) provides detailed insights that facilitate accurate tumor targeting and effective treatment planning. However, challenges such as spill-out can distort tumor size, which complicates accurate segmentation. This study introduces a Paired Multi-Scale Attention Network (P-MANet) using Top-Hat Cross-Features to address these challenges by enhancing the identification of true positives while minimizing false positives in liver tumor segmentation. P-MANet employs a dual-branch architecture. The first branch utilizes a Multi-Scale Attention Network (MA-Net) to process SPECT/CT datasets, whereas the second branch applies the White Top-Hat Transform to extract features that are then integrated with those from the first branch. This innovative approach effectively mitigates errors stemming from spectral light variation, leading to improved accuracy in tumor delineation. The model was trained on a dataset comprising 43 cases of 99mTc-MAA SPECT/CT. P-MANet achieved a Dice Similarity Coefficient (DSC) of 67.93% and 66.56% for normal and abnormal spectral light distributions, respectively, and obtained a DSC of 67.00%, on average, which outperformed other models, including results from a previous study that was tested on the same dataset.
- 18 February 2025
- Published by Elsevier in Journal of Food Engineering
- Vol. 395
Tuna defect classification and grading using Twins transformer
- 18 February 2025
- Published by Elsevier in Journal of Food Engineering
- Vol. 395
Tuna defect classification and grading using Twins transformer
- Open Access
- 7 January 2025
- Published by MDPI AG in Journal of Imaging
- Vol. 11 (1) , 13
Combined Input Deep Learning Pipeline for Embryo Selection for In Vitro Fertilization Using Light Microscopic Images and Additional Features
The current process of embryo selection in in vitro fertilization is based on morphological criteria; embryos are manually evaluated by embryologists under subjective assessment. In this study, a deep learning-based pipeline was developed to classify the viability of embryos using combined inputs, including microscopic images of embryos and additional features, such as patient age and developed pseudo-features, including a continuous interpretation of Istanbul grading scores by predicting the embryo stage, inner cell mass, and trophectoderm. For viability prediction, convolution-based transferred learning models were employed, multiple pretrained models were compared, and image preprocessing techniques and hyperparameter optimization via Optuna were utilized. In addition, a custom weight was trained using a self-supervised learning framework known as the Simple Framework for Contrastive Learning of Visual Representations (SimCLR) in cooperation with generated images using generative adversarial networks (GANs). The best model was developed from the EfficientNet-B0 model using preprocessed images combined with pseudo-features generated using separate EfficientNet-B0 models, and optimized by Optuna to tune the hyperparameters of the models. The designed model’s F1 score, accuracy, sensitivity, and area under curve (AUC) were 65.02%, 69.04%, 56.76%, and 66.98%, respectively. This study also showed an advantage in accuracy and a similar AUC when compared with the recent ensemble method.
- 7 January 2025
- Published by MDPI AG in Journal of Imaging
- Vol. 11 (1) , 13
Open AccessCombined Input Deep Learning Pipeline for Embryo Selection for In Vitro Fertilization Using Light Microscopic Images and Additional Features
The current process of embryo selection in in vitro fertilization is based on morphological criteria; embryos are manually evaluated by embryologists under subjective assessment. In this study, a deep learning-based pipeline was developed to classify the viability of embryos using combined inputs, including microscopic images of embryos and additional features, such as patient age and developed pseudo-features, including a continuous interpretation of Istanbul grading scores by predicting the embryo stage, inner cell mass, and trophectoderm. For viability prediction, convolution-based transferred learning models were employed, multiple pretrained models were compared, and image preprocessing techniques and hyperparameter optimization via Optuna were utilized. In addition, a custom weight was trained using a self-supervised learning framework known as the Simple Framework for Contrastive Learning of Visual Representations (SimCLR) in cooperation with generated images using generative adversarial networks (GANs). The best model was developed from the EfficientNet-B0 model using preprocessed images combined with pseudo-features generated using separate EfficientNet-B0 models, and optimized by Optuna to tune the hyperparameters of the models. The designed model’s F1 score, accuracy, sensitivity, and area under curve (AUC) were 65.02%, 69.04%, 56.76%, and 66.98%, respectively. This study also showed an advantage in accuracy and a similar AUC when compared with the recent ensemble method.
- Open Access
- 31 January 2024
- Published by Wiley in Journal of Applied Clinical Medical Physics
- Vol. 25 (3) , e14283
Automated size‐specific dose estimates framework in thoracic CT using convolutional neural network based on U‐Net model
Purpose: This study aimed to develop an automated method that uses a convolutional neural network (CNN) for calculating size‐specific dose estimates (SSDEs) based on the corrected effective diameter (Deffcorr) in thoracic computed tomography (CT).Methods: Transaxial images obtained from 108 adult patients who underwent non‐contrast thoracic CT scans were analyzed. To calculate the Deffcorr according to Mihailidis et al., the average relative electron densities for lung, bone, and other tissues were used to correct the lateral and anterior–posterior dimensions. The CNN architecture based on the U‐Net algorithm was used for automated segmentation of three classes of tissues and the background region to calculate dimensions and Deffcorr values. Then, 108 thoracic CT images and generated segmentation masks were used for network training. The water‐equivalent diameter (Dw) was determined according to the American Association of Physicists in Medicine Task Group 220. Linear regression and Bland–Altman analysis were performed to determine the correlations between SSDEDeffcorr(automated), SSDEDeffcorr(manual), and SSDEDw.Results: High agreement was obtained between the manual and automated methods for calculating the Deffcorr SSDE. The mean values for the SSDEDeffcorr(manual), SSDEDw, and SSDEDeffcorr(automated) were 14.3 ± 2.1 mGy, 14.6 ± 2.2 mGy, and 14.5 ± 2.4 mGy, respectively. The U‐Net model was successfully trained and used to accurately predict SSDEs, with results comparable to manual‐labeling results.Conclusion: The proposed automated framework using a CNN offers a reliable and efficient solution for determining the Deffcorr SSDE in thoracic CT.
- 31 January 2024
- Published by Wiley in Journal of Applied Clinical Medical Physics
- Vol. 25 (3) , e14283
Open AccessAutomated size‐specific dose estimates framework in thoracic CT using convolutional neural network based on U‐Net model
Purpose: This study aimed to develop an automated method that uses a convolutional neural network (CNN) for calculating size‐specific dose estimates (SSDEs) based on the corrected effective diameter (Deffcorr) in thoracic computed tomography (CT).Methods: Transaxial images obtained from 108 adult patients who underwent non‐contrast thoracic CT scans were analyzed. To calculate the Deffcorr according to Mihailidis et al., the average relative electron densities for lung, bone, and other tissues were used to correct the lateral and anterior–posterior dimensions. The CNN architecture based on the U‐Net algorithm was used for automated segmentation of three classes of tissues and the background region to calculate dimensions and Deffcorr values. Then, 108 thoracic CT images and generated segmentation masks were used for network training. The water‐equivalent diameter (Dw) was determined according to the American Association of Physicists in Medicine Task Group 220. Linear regression and Bland–Altman analysis were performed to determine the correlations between SSDEDeffcorr(automated), SSDEDeffcorr(manual), and SSDEDw.Results: High agreement was obtained between the manual and automated methods for calculating the Deffcorr SSDE. The mean values for the SSDEDeffcorr(manual), SSDEDw, and SSDEDeffcorr(automated) were 14.3 ± 2.1 mGy, 14.6 ± 2.2 mGy, and 14.5 ± 2.4 mGy, respectively. The U‐Net model was successfully trained and used to accurately predict SSDEs, with results comparable to manual‐labeling results.Conclusion: The proposed automated framework using a CNN offers a reliable and efficient solution for determining the Deffcorr SSDE in thoracic CT.
- Open Access
- 13 October 2022
- Published by Elsevier in Smart Agricultural Technology
- Vol. 3
Fruit maturity grading framework for small dataset using single image multi-object sampling and Mask R-CNN
- 13 October 2022
- Published by Elsevier in Smart Agricultural Technology
- Vol. 3
Open AccessFruit maturity grading framework for small dataset using single image multi-object sampling and Mask R-CNN