This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 6.8k
mx.contrib.tensorrt.tensorrt_bind got a bad result! #12598
Labels
Comments
Thanks for submitting the issue @lizhen2017 |
@KellenSunderland - can you take a look please? |
@lizhen2017 Thanks for the report. Would you be able to share the model you're working with? |
@KellenSunderland @KellenSunderland I implement a tiny-yolov3 network, the network inference well without trt api, and how can share the model? |
@KellenSunderland @larroy bouncing... |
@mxnet-label-bot [TensorRT] |
@lizhen2017 Sorry missed your question. If you're using the symbol API you can save using https://mxnet.apache.org/api/python/symbol/symbol.html#mxnet.symbol.Symbol.save |
5 tasks
5 tasks
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Environment:mxnet-tensorrt-cu90 tensorrt 4.0, cuda9.0, cudnn 7.1 python3.5
(1)Inference with sym.simple_bind got right detection result.
executor = sym.simple_bind(ctx=ctx, data=batch_shape, grad_req='null', force_rebind=True)
executor.copy_params_from(arg_params, aux_params)
(2) Inference with mx.contrib.tensorrt.tensorrt_bind got bad result
os.environ['MXNET_USE_TENSORRT'] = '1'
arg_params.update(aux_params)
all_params = dict([(k, v.as_in_context(mx.gpu(1))) for k, v in arg_params.items()])
executor = mx.contrib.tensorrt.tensorrt_bind(sym, ctx=mx.gpu(1), all_params=all_params,
data=batch_shape, grad_req='null', force_rebind=True)
(3)Inference with gluon got right detection result.
self.__model = gluon.nn.SymbolBlock(outputs=mx.sym.load(self.__symbol), inputs=mx.sym.var('data'))
self.__model.load_parameters(self.__params, ctx=self.__ctx)
I don't know why.Thanks for your help! Will using tensort makes a lower accuracy?
The text was updated successfully, but these errors were encountered: