Conceptual Cognitive Modeling for Fine-Grained Annotation Quality Assessment of Object Detection Datasets

Table 9

Comparison of detection results based on the original training dataset and downsampled dataset (the average values are computed for the twenty classes).

Class	Object number in the training dataset	mAP (original)	mAP (downsampled)
Person	5447	0.779	0.778
Car	1644	0.831	0.807
Chair	1432	0.520	0.511
Bottle	634	0.576	0.519
Potted plant	625	0.459	0.376
Cow	356	0.767	0.721
Dining table	310	0.682	0.671
Bus	272	0.772	0.776
On average	783.1	0.714	0.678