Visualizing Defects in Amazon’s ARMBench Dataset Using Embeddings and OpenAI’s CLIP Model

In this blog post, we explored Amazon's recently released computer vision dataset for training "pick and place" robots using the open-source FiftyOne toolset. The ARMBench dataset is the largest computer vision dataset ever captured in an industrial product-sorting setting, featuring over 235,000 pick and place activities on 190,000 objects. We focused on the Image Defect Detection subset of data and utilized FiftyOne to visualize it. Additionally, we created embeddings with the OpenAI CLIP model to explore defects further. The practical applications of "pick and place" robots include manufacturing, packaging, sorting, and inspection tasks that require speed and accuracy.