Keras-yolo3: How to implement the SoftNMS algorithm instead of the original tf.image.non_maximum_suppression(). 如何实现使用softnms 来代替原先的tf.image.non_maximum_suppression()函数

[DocF]

Hi，everyone！
I have been confused by a problem recently. How to implement the SoftNMS algorithm instead of the original tf.image.non_maximum_suppression(). Here is the SoftNMS repo: https://github.com/DocF/Soft-NMS . (这里是SoftNMS的代码仓库)
I test it successfully when the boxes and scores are np.ndarray. But I get back this big error message errors when instead of the original tf.image.non_maximum_suppression().

[S601327412]

You can find Tensorflow C++ source code,tf.image.non_maximum_suppression() is softnms actually。

[Stinky-Tofu]

Hi，everyone！
I have been confused by a problem recently. How to implement the SoftNMS algorithm instead of the original tf.image.non_maximum_suppression(). Here is the SoftNMS repo: https://github.com/DocF/Soft-NMS . (这里是SoftNMS的代码仓库)
I test it successfully when the boxes and scores are np.ndarray. But I get back this big error message errors when instead of the original tf.image.non_maximum_suppression().

You can reference my code https://github.com/Stinky-Tofu/Stronger-yolo

[jamiechoi1995]

tf.image.non_maximum_suppression

@S601327412 non_maximum_suppression() refers to different NMS versions(V2, V3, V4),
I read the code of these versions, they all finally use DoNonMaxSuppressionOp(),
however, I don't think it uses soft-NMS.

see: https://github.com/tensorflow/tensorflow/blob/a26413ef0a179dd93b5228b031de83dce97a8684/tensorflow/core/kernels/non_max_suppression_op.cc#L155

[zpge]

mark and waiting for solutions.

[TianwenZhang0825]

你好，同学，请问这个soft-nms问题得到解决了吗？我这几天一直在弄。