The quality of germinated oil palm seeds directly affects the growth and yield of oil palm industry. Therefore, strict filtering process is required before dispatching the germinated oil palm seeds to the market. Manual inspection of individual germinated oil palm seeds is laborious and error-prone. We propose a high-throughput system that can accurately compute the correspondences of arbitrary number of oil palm seeds in multi-view images captured from unknown camera positions, and subsequently predict the quality of individual oil palm seeds using a classification network. First, an object detection model is fine-tuned to detect the oil palm seeds in each image. Then, the transformation model between each image pair is estimated. We introduce an abstract representation to address the wide baseline problem and challenging visual information, which results in improved accuracy of the transformation model estimation. Based on the seed candidates detected by the object detection model and the estimated transformation models, the correspondences of oil palm seeds can be computed. We propose a multi-view oil palm seed quality classification network to predict the quality of each oil palm seed in the set. The classification network receives the multi-view images of an individual oil palm seed as input, and output the probability of the input oil palm seed being a bad seed. The proposed multi-view oil palm seed quality classification model is able to achieve an accuracy of 90%. The proposed system is able to process a large batch of oil palm seeds simultaneously using a single set of multi-view images, without needing to acquire multi-view images for each individual oil palm seed. This not only reduces laborious effort in data acquisition, but also highly reduces the memory consumption. •High-throughput oil palm seeds quality classification.•Correspondence matching of oil palm seeds with unknown camera positions.•Quality classification of germinated oil palm seed based on multi-view images.