Jianqiang Zhou

Applied data analysis techniques to enhance training datasets and gain deeper insights into model performance, leveraging expertise in image and color science. Trained and fine-tuned multi-task machine learning models with architectures such as ResNet, ViT, and Swin Transformer. Deployed machine learning models efficiently using data versioning (Git+DVC), experiment tracking (MLflow), and scalable APIs (FastAPI). Specialized in training image quality models with mixed-bit-depth images (HDR and SDR), optimizing performance for real-world viewing conditions.

- Trained several high-accuracy classification models (data preparation, annotation guidelines, annotation, model selection&training&fine-tune, and performance evaluation). - Trained several segmentation models (shadow detection, foreground-background separation). - Designed a solution for improving the text/watermark detection model. - Pilot owner: The product owner of experimental projects follows the client’s requirements.