METU uses FiftyOne for cutting-edge deep learning and computer vision research
The Deep Learning and Computer Vision group at METU conducts research on the cutting-edge topics of deep learning and computer vision. Their research is focused on object detection, adversarial attacks, medical imaging, 3D model generation, generative models, and GPU programming.
METU’s success story with FiftyOne
METU collaborates with the Department of Gastroenterology at Marmara University, Turkey to conduct medical research on inflammatory bowel disease (IBD). One of the main challenges in IBD diagnosis is the high interobserver and intraobserver variability among the assessments of the doctors. With the aid of deep learning, real-time and objective feedback can be provided to doctors during colonoscopy operations.
METU uses FiftyOne at various stages of research. First, they make use of FiftyOne’s CVAT integration to request and load annotations from doctors. Next, they compare doctors’ annotations and their models’ predictions. Then, using FiftyOne Brain’s embedding space visualization tools (t-SNE specifically), they get a better understanding of their data distribution. In a branch of this research, they produce synthetic images with GANs and again, refer to FiftyOne Brain tools for help identifying the most informative/unique images by eliminating “near duplicates”. In the coming months, doctors will start sending videos and METU will start to make use of FiftyOne’s video analysis capabilities.
Workflows & highlights
CVAT
integration
Model
evaluation
Embeddings
visualizations
Image
uniqueness
