Deep learning researcher specializing in computer vision with a proven track record of developing cutting-edge models for object detection, image segmentation, and anomaly detection in CCTV driving videos. Proficient in PyTorch, TensorFlow, and Python, with expertise in creating DeepStream plugins for seamless integration of deep learning algorithms. Trusted by management to lead research initiatives and ensure team effectiveness and efficiency.
1. Developed a new model , the TempoLearn network, which outperformed existing state-of-the-art models by achieving a 16.5% higher incident localization score (AUC) on the largest and most complex Detection of Traffic Anomaly (DoTA) dataset. Traffic accident datasets are available.
2. We demonstrate the effectiveness of the TempoLearn network through experiments on the Car Crash Dataset (CCD) and further validate its robustness and performance on different datasets.
3. Demonstrate proficiency in advanced machine learning methodologies by leveraging multimodal learning techniques to improve model accuracy and generalization capabilities.
https://ieeexplore.ieee.org/document/10360840
2022 13th International Conference on Information and Communication Technology Convergence (ICTC)
1. Demonstrated a state-of-the-art multimodal learning approach, integrating the convolutional learning method with a self-attention model to extract relevant features and model temporal occurrences.
2. Demonstrated the effectiveness of the proposed architecture in accurately classifying segmented traffic accidents into predefined groups and highlighted its practical applicability in improving road safety and accident prevention.
3. Provided valuable insights and recommendations for future research and implementation, emphasizing the commitment to continuous improvement and advancement in the field of AI-based traffic analysis and anomaly detection.
https://ieeexplore.ieee.org/document/9952473
2021 21st International Conference on Control, Automation and Systems (ICCAS)
1. Developed a new approach to video captioning that incorporates important contextual information such as human actions, objects, location, time, and weather.
2. Created a new dataset that supplements the existing MSVD dataset with environmental context labels, addressing a significant gap in prior video captioning research.
3. Implemented a state-of-the-art machine learning model that combines R(2+1)D 3D CNN for video feature extraction with S2VT RNN for environmental information encoding and decoding.
4. Adopted a sequence-to-sequence framework tailored for video analysis to ensure the model generates coherent and contextually appropriate captions.
5. Validated that our model delivers competitive performance compared to existing video captioning models evaluating with BLEU, METEOR, ROUGE-L, and CIDEr.
https://ieeexplore.ieee.org/document/9649840