> 'data' in tvm can be composed by the 'boxes' and 'score' in tf/pt. > valid_counts is to support different ways to generate valid number of boxes, > as we have discussed. We don't need score_threshold in this api then. > The extra argument 'return_indices' is to decide whether to return box > indices. I put it right after the input tensor since it is mandatory. > Other additional arguments make tvm implementation the superset of tf/pt > implementation. When converting tf/pt nms, we need to set > > ``` > force_suppress=True, > topk=-1, > id_index=0, > invalid_to_bottom=False > ```
Hi, I saw you mentioned that when users use tf/pt models with tvm's non_max_suppression op, they can compose 'boxes' and 'score' to generate the 'data' parameter. Could you give me some hints for how to compose 'boxes' and 'scores'? -- You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub: https://github.com/dmlc/tvm/issues/2535#issuecomment-515648299
