Implement and improve state-of-the-art computer vision models.Provide multimodal tuning solutions (e.g., PaLI, Gemini) for application teams.Highlight works:Support object detection model (YOLOv7) and panoptic segmentation models (panoptic UViT, MaskConver) in TF Model Garden.Implement LoRA for PaLI, and conduct excessive research on LoRA layer effects.Provide downstream full and LoRA fine-tuning capabilities for Gemini multimodal models, and works are adopted by multiple application teams.

Supporting machine learning infrastructure for Geo-specific applications.Build reusable training, eval and inference pipelines to launch Tensorflow models.