search

ddonco / retinanet-pytorch

GNU General Public License v3.0
4 stars 1 forks source link

Parameter check failed at: ../builder/Network.cpp::addConcatenation::308, condition: nbInputs > 0 && nbInputs < MAX_CONCAT_INPUTS #1

Open lucasjinreal opened 4 years ago

lucasjinreal commented 4 years ago 
 $ retinanet export weights/retinanet_rn50fpn_4gpus.pth  model_fp16.plan --batch 1 --size 1280
Loading model from retinanet_rn50fpn_4gpus.pth...
model: ['ResNet50FPN'] max stride is: 128
     model: RetinaNet
  backbone: ResNet50FPN
   classes: 80, anchors: 9
command: export
Exporting to ONNX...
Building FP16 core model...
Building accelerated plugins...
Parameter check failed at: ../builder/Network.cpp::addConcatenation::308, condition: nbInputs > 0 && nbInputs < MAX_CONCAT_INPUTS
[1]    13640 segmentation fault  retinanet export weights/retinanet_rn50fpn_4gpus.pth model_fp16.plan --batch

Do u got this issue ? TensorRT 6

ddonco commented 4 years ago

I use this docker image provided by nvidia to train, inference and export, and I have no problems with exporting: vcr.io/nvidia/pytorch:19.09-py3

I believe this image uses TensorRT 6 last I checked.