Documentation for non_max_suppression

Questions
#1

After reading discussion and documentation about vision.non_max_suppression(), I still don’t understand what the argument valid_counts is about.
It’s a 1D tensor, but how long?
And what do the entries actually signify?

#2

This tensor is the valid number of boxes in each batch sample. So the length is the same as batch size.

#3

Hi, Which frontends do support NMS in TVM?

#4

MXNet for now. TF in progress.

#5

Sorry to bother you, and I would like to know when TF will be supported @ kevinthesun

1 Like
#6

I am working on tf nms support, wrote an draft version. Will pr it after a few fixes. @ydy

#7

sounds great! thanks so much.

#8

Sorry to bother you, but are you interested in supporting “Assert”?I’m a little confused because it’s more like a type than an OP, so I don’t know how to support it. @yongwww

#9

@ydy Assert is in our queue, will work on adding the support. One option is to add a general assert in relay, the other option is to assemble the op with other ops

1 Like