mmdetection model Faster_rcnn failed with pplnn

[Maosquerade]

mmdetection model Faster_rcnn failed with pplnn

[openppl-public]

I followed the document of converting model with openmmp and generated pretrained faster_rcnn model , but when I used x86 pplnn to run this model, there is an error says:

[INFO][2021-07-06 08:48:49.498][pplnn.cc:683] ppl.nn version: 7dd75a1077867fc9a762449953417088446ae2f8-dirty
[INFO][2021-07-06 08:48:49.498][pplnn.cc:110] ***** register X86Engine *****
[INFO][2021-07-06 08:48:49.761][simple_graph_partitioner.cc:90] total partition(s) of graph[torch-jit-export]: 1.
[ERROR][2021-07-06 08:48:50.556][kernel.cc:14] reshape kernel[Expand_1100] failed: invalid value
[ERROR][2021-07-06 08:48:50.556][kernel.cc:47] BeforeExecute() of kernel[Expand_1100] failed: invalid value
[ERROR][2021-07-06 08:48:50.556][scheduler_common.cc:153] exec kernel[Expand_1100] failed: invalid value
[ERROR][2021-07-06 08:48:50.556][sequential_scheduler.cc:99] execute kernel[Expand_1100] failed: invalid value
[ERROR][2021-07-06 08:48:50.556][pplnn.cc:784] Run() failed: invalid value

The mobilenet can execute successfully. Can I do anything to make this model execute right?

Can you post the command sequence that are used to convert the faster-rcnn model?

[Maosquerade]

My command is just as the insturction of your model-convert-guide , the Example: Converting Faster R-CNN :

cd mmdetection && mkdir checkpoints && cd checkpoints
wget https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth

python ../tools/deployment/pytorch2onnx.py ../configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py \
faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth \
--output-file faster_rcnn.onnx --simplify --dynamic-export

[openppl-public]

My command is just as the insturction of your model-convert-guide , the Example: Converting Faster R-CNN :

cd mmdetection && mkdir checkpoints && cd checkpoints
wget https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth

python ../tools/deployment/pytorch2onnx.py ../configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py \
faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth \
--output-file faster_rcnn.onnx --simplify --dynamic-export

We cannot reproduce your problem. Can you send your converted model to [email protected]?

[Maosquerade]

Thanks for relpying! I have sent the model and my environment to your email.

[ZichenTian]

Thanks for relpying! I have sent the model and my environment to your email.

We have successfully reproduced your problem on your model. The bug has been fixed and we merged a PR:
#15

By our test, your model now can be infered successfully, and its result is same as onnxruntime.
Please have a try again.

[Maosquerade]

That worked! Thanks a lot!