Open htyj123 opened 3 years ago
It is relatively simple to modify the code for MaxBoxAcc
to compute the top-1 localization. MaxBoxAcc
determines the correctness of the box (with a given iou threshold) for each image. We can modify the determination of correctness by additionally taking the classification correctness into account. Then we have the top-1 localization accuracy.
It is not as simple for the PxAP
metric since the metric is computed pixel-wise, rather than image-wise.
Can I directly use the model trained by your code to test the top-1 localization ? and could you please provide the implement of the top-1 localization evaluation. Thank you