New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
How I know the searched people is the same as the people in the gallery? #6
Comments
|
After the function _compute_iou in the https://github.com/serend1p1ty/SeqNet/blob/58471df02baacd338a53037e8856bc05b2be2229/eval_func.py#L246,the model think the j-th detection is correct,can we multiply sim score by 2 to highlight the j-th detection? I have encountered such a scenario which sim socre of the overall detection result is low,while the sim score of the wrong candidate is higher than that of the correct one.Rank1 is improved by doubling sim score of j-th detection. |
Top-1 can be improved by multiplying the similarity of the j-th detection by 2 when you know that the j-th detection is ground truth, but it is a "cheating" behavior. In addition, multiplying by a larger number will further improve Top-1. For fair comparison, the evaluation rule CAN NOT be modified. |
i want to finetune the second part of the seqnet model with custom data, while freezing the first part, detection.how can i do that? By setting the requires_grad attribute of the parameter of backbone and rpn of model is False? |
a person may appear in serveral pictures,and the dataset has the unique person ID for every one?
because gallery and query has duplicate pics, and one pic may contain serveral persons, so you compute IOU between searched perosn box with groud truth to get positive case?
thank you!
The text was updated successfully, but these errors were encountered: