FDIM: A Feature-distance-based Generic Video Quality Metric for Versatile Codecs
Researchers have introduced FDIM, a new video quality metric designed to evaluate both traditional and neural video codecs across Standard Dynamic Range (SDR) and High Dynamic Range (HDR) content. FDIM utilizes a hybrid approach, combining deep learning for multi-scale feature extraction with hand-crafted features to capture a wide range of distortions. Trained on a large dataset of over 16,000 video sequences, FDIM has demonstrated strong generalization capabilities across various codecs and content types, correlating well with subjective quality assessments. AI
IMPACT Provides a generalized metric for evaluating neural video codecs, potentially accelerating research and development in video compression.