Embedded Files
We perform a systematic evaluation of twelve representative object detectors covering three different categories of architectures (i.e., two-stage, one-stage, transformer architectures). The evaluation use our sixteen real-world image corruptions, each of which contained five severity levels. The database contains coco and bdd100k.
The systematic evaluation results is shown in this tabel. The number in the table is average CE(corruption error) of coco and bdd100k. The CE of every database is average CE of five severity levels.
In addition to, we also record every severity level data that the CE(corruption error) of object detectors models. The number (i.e.,1,2,3,4,5) in the third row indicates the severity level, and the severity levels to add as the number gets bigger.
The dataset is COCO.
The Camera Damage CE(corruption error) of object detectors models.
The Image Signal Processor Damage CE(corruption error) of object detectors models.
The Board-level Damage CE(corruption error) of object detectors models.
The dataset is BDD100K.
The Camera Damage CE(corruption error) of object detectors models.
The Image Signal Processor Damage CE(corruption error) of object detectors models.
The Board-level Damage CE(corruption error) of object detectors models.
Page updated
Google Sites
Report abuse