Micro Yi's collaboration with East China Normal University has resulted in their paper, 'FSDC-DETR: A Frequency-Spatial Domain Collaborative DETR for Small Object Detection,' being accepted at ECCV 2026. This prestigious conference received 10,473 submissions, with only 2,883 accepted, highlighting the significance of their research in the field of computer vision.
The FSDC-DETR framework addresses the challenge of detecting tiny defects in precision manufacturing, such as bubbles and micro-cracks, which often occupy only a few dozen pixels in high-resolution images. By integrating CNN's local high-frequency detail extraction with ViT's global low-frequency modeling, the architecture effectively retains edge and texture information, significantly reducing both missed and false detections.
Currently, FSDC-DETR is embedded in Micro Yi's industrial embodied intelligent quality inspection robots, achieving a cold start detection rate of 95.55% in complex scenarios like die-casting. With over 23TB of aligned precision data and substantial R&D investment, Micro Yi is positioned to lead the embodied intelligence market, projected to grow significantly by 2026.
Editor's Note
The acceptance of Micro Yi's paper at ECCV 2026 underscores the growing importance of advanced AI technologies in industrial applications. As the market for embodied intelligence expands, companies that can effectively bridge the gap between technological validation and large-scale deployment will gain a competitive edge. The integration of cutting-edge research into practical solutions is crucial for the evolution of intelligent manufacturing.
Leave a comment