It is all well and good to use mAP as a metric for the COCO and Pascal datasets. Those datasets are somewhat well distributed amongst the classes. However, more often than not, the real-world dataset will be highly skewed towards one or two classes.
In my case, my dataset has a lot of examples for Cars and Trucks. And very few for Person, Bicycle, Motorcycle, Bus.
Therefore it is needed to see the Precision and Recalls for the classes Cars and Trucks.
Any idea how to plot individual ROC curves for the classes Cars and Trucks using COCOEval?
Hello,
It is all well and good to use mAP as a metric for the COCO and Pascal datasets. Those datasets are somewhat well distributed amongst the classes. However, more often than not, the real-world dataset will be highly skewed towards one or two classes.
In my case, my dataset has a lot of examples for Cars and Trucks. And very few for Person, Bicycle, Motorcycle, Bus. Therefore it is needed to see the Precision and Recalls for the classes Cars and Trucks.
Any idea how to plot individual ROC curves for the classes Cars and Trucks using COCOEval?