Error in running movinet-pytorch to onnx converted model about movinet-pytorch HOT 4 CLOSED

I would suggest you avoid screenshots, the code should be reported in text. The same is true also for the errors, in fact the error messages are also incomplete.
This may help onnx/onnx#3192.
You could try to convert the model with pretrained = False and tf_like = False. If it works there could be problems related to the fact that this architecture emulates tensorflow models.

from movinet-pytorch.

hi @Atze00 I did a mistake of saving model as it is in torch.save meaning torch.save(model) instead of recommended way of torch.save(model.state_dict(),PATH) which led to this all nonsense

1. onnx converted and excuted with out error

2. I have yet to run sanity tests like onnx output and this repo output is matching on some videos

3. Now i am proceeding to export this pytorch model to tensorrt , there are some hiccups like oom

Thanks @Atze00 your repo model implementation is giving satisfactory results on our use case

from movinet-pytorch.

Glad to hear that!

from movinet-pytorch.

@Atze00 i am closing this ticket as its purpose is served , just one request is there any fundamental way to speed up bare pytorch on jetson for movinets , i tried other options but they are giving same speed as bare pytorch.

Anything from your experience would be a lot helpful than u know

from movinet-pytorch.