模型转换nms算子报错

[color=rgba(0, 0, 0, 0.85)][backcolor=rgba(0, 0, 0, 0.04)]像模型中有nms操作，转换报错的情况，有什么解决办法吗，是不是只能把nms给分割出来进行后处理这一个办法
[color=rgba(0, 0, 0, 0.85)][backcolor=rgba(0, 0, 0, 0.04)]E build: The input 0 of NonMaxSuppression('/NonMaxSuppression') need to be constant! It will cause the graph to be a dynamic graph!                  Remove NonMaxSuppression('/NonMaxSuppression') manually and try again!W build: ===================== WARN(1) =====================E rknn-toolkit2 version: 2.3.0E build: Traceback (most recent call last):  File "rknn/api/rknn_log.py", line 344, in rknn.api.rknn_log.error_catch_decorator.error_catch_wrapper  File "rknn/api/rknn_base.py", line 1945, in rknn.api.rknn_base.RKNNBase.build  File "rknn/api/graph_optimizer.py", line 893, in rknn.api.graph_optimizer.GraphOptimizer.fold_constant  File "rknn/api/graph_optimizer.py", line 366, in rknn.api.graph_optimizer._dynamic_check  File "rknn/api/rknn_log.py", line 95, in rknn.api.rknn_log.RKNNLog.eValueError: The input 0 of NonMaxSuppression('/NonMaxSuppression') need to be constant! It will cause the graph to be a dynamic graph!                  Remove NonMaxSuppression('/NonMaxSuppression') manually and try again!I ===================== WARN(0) =====================E rknn-toolkit2 version: 2.3.0Traceback (most recent call last):  File "rknn/api/rknn_log.py", line 344, in rknn.api.rknn_log.error_catch_decorator.error_catch_wrapper  File "rknn/api/rknn_base.py", line 1945, in rknn.api.rknn_base.RKNNBase.build  File "rknn/api/graph_optimizer.py", line 893, in rknn.api.graph_optimizer.GraphOptimizer.fold_constant  File "rknn/api/graph_optimizer.py", line 366, in rknn.api.graph_optimizer._dynamic_check  File "rknn/api/rknn_log.py", line 95, in rknn.api.rknn_log.RKNNLog.eValueError: The input 0 of NonMaxSuppression('/NonMaxSuppression') need to be constant! It will cause the graph to be a dynamic graph!                  Remove NonMaxSuppression('/NonMaxSuppression') manually and try again!During handling of the above exception, another exception occurred:Traceback (most recent call last):  File "/home/lly/rknn_model_zoo-main/examples/yolov5/python/convert.py", line 62, in <module>    ret = rknn.build(do_quantization=do_quant, dataset=DATASET_PATH)  File "/home/lly/conda/miniconda3/envs/py310/lib/python3.10/site-packages/rknn/api/rknn.py", line 192, in build    return self.rknn_base.build(do_quantization=do_quantization, dataset=dataset, expand_batch_size=rknn_batch_size)  File "rknn/api/rknn_log.py", line 349, in rknn.api.rknn_log.error_catch_decorator.error_catch_wrapper  File "rknn/api/rknn_log.py", line 95, in rknn.api.rknn_log.RKNNLog.eValueError: Traceback (most recent call last):  File "rknn/api/rknn_log.py", line 344, in rknn.api.rknn_log.error_catch_decorator.error_catch_wrapper  File "rknn/api/rknn_base.py", line 1945, in rknn.api.rknn_base.RKNNBase.build  File "rknn/api/graph_optimizer.py", line 893, in rknn.api.graph_optimizer.GraphOptimizer.fold_constant  File "rknn/api/graph_optimizer.py", line 366, in rknn.api.graph_optimizer._dynamic_check  File "rknn/api/rknn_log.py", line 95, in rknn.api.rknn_log.RKNNLog.eValueError: The input 0 of NonMaxSuppression('/NonMaxSuppression') need to be constant! It will cause the graph to be a dynamic graph!                  Remove NonMaxSuppression('/NonMaxSuppression') manually and try again!

是的，非标准算子肯定是要单独拉出来做的。
你也可以试下先转onnx再转rknn，onnx能把他拆成标准的多个子算子就可以转

jefferyzhang 发表于 2025-10-30 17:03
是的，非标准算子肯定是要单独拉出来做的。
你也可以试下先转onnx再转rknn，onnx能把他拆成标准的多个子算 ...

这个已经是onnx模型了