Intellindust-AI-Lab
Too high false positives
I have tried dfine and deim and found that although they have a high detection rate, their false detection rate is also absurdly high, especially when the same object has almost no ability to distinguish between different states and outputs both states. Although the detection rate is high, it is not what I want. How to adjust the balance between detection rate and false detection rate
我试过 dfine 和 deim，发现虽然它们检出率很高，但误检率也高得离谱，尤其是同一个物体几乎没有区分不同状态的能力，同时输出两种状态。虽然检出率很高，但这不是我想要的。如何调整检测率和误检率之间的平衡
I'm also experiencing this problem with excessive false positives. May I ask if you have tried some solutions, such as making the perturbation not limited to the box neighborhood in the denoising perturbation. We want the model to learn more background class information.
Hey have you all tried reducing the num_queries parameter? The default is 300 and if you are training on a custom dataset that has less objects per image, this may be too many.
Hey have you all tried reducing the
num_queriesparameter? The default is 300 and if you are training on a custom dataset that has less objects per image, this may be too many.
Have you tried to reduce the num_queries parameter? Does is work to make FP lower?
