Supplementary for Mixed Supervised Object Detection by Transferring Mask Prior and Semantic Similarity Y an Liu

In this supplementary material, we will provide more analyses of mask prior in Section 1 and similarity transfer in Section 2. We will show the visualization results in Section 3 and the performance variance Figure 2) to better investigate the effectiveness of mask prior. We treat the similarity prediction task as a binary classification task, in which the binary label 1 ( resp. For VOC test set, we repeat the above procedure. The precision, recall and F1 scores are summarized in Table 1. We observe that the gap between the performance of similarity network on base categories and novel categories is negligible ( e.g., F1 Scores " person " and calculate their pairwise semantic similarity scores by applying the trained similarity It is well worth noting that the average similarity score will be affected slightly by the number of outliers if batch size increases to a large scale ( e.g., 64).