A simple and effective way to improve long-tailed object detection (L TOD) is to use extra data to increase the training samples for tail classes. However, collecting bounding box annotations, especially for rare categories, is costly and tedious. Therefore, previous studies resort to datasets with image-level labels to enrich the amount of samples for rare classes by exploring image-level semantics (as shown in Figure 1 (a)). While appealing, directly learning from such data to benefit detection is challenging since they lack bounding box annotations that are essential for object detection.

During the inference, we set the background score as 0.95.

This paper studies the problem of weakly open-vocabulary semantic segmentation (WOVSS), which learns to segment objects of arbitrary classes using mere image-text pairs. Existing works turn to enhance the vanilla vision transformer by introducing explicit grouping recognition, i.e., employing several group tokens/centroids to cluster the image tokens and perform the group-text alignment. Nevertheless, these methods suffer from a granularity inconsistency regarding the usage of group tokens, which are aligned in the all-to-one v.s.

Major challenges in scaling self-training are the choice of label space, pseudo-annotation filtering, and training efficiency.

Photorealistic generation is undergoing a period that future scholars may well compare to the enlightenment era.

The resulting Hummingbird model, suitably prompted, performs various scene understanding tasks without modification while approaching the performance of specialists that have been finetuned for each task.

Through extensive experiments on prevalent image datasets, we demonstrate the superiority of HASSOD over existing methods, thereby advancing the state of the art in self-supervised object detection. Notably, we improve Mask AR from 20.2 to 22.5 on L VIS, and from 17.0 to 26.0 on SA-1B.