Ghazaleh

Sorry I noticed the models are pretrained on kinetics not hmdb. Still I wasn't expecting such a low accuracy.

> I'm wondering if I was running into the same issue. > > When connecting to a random device with the pypylon sample code such as `grab.py` I was able...

@SMA2016a thanks you for your response. I was trying to use a VPN tunnel to connect to the camera and grab the frames with pypylon. I finally managed to establish...

@PolymaTh-EE98 thank you for your response. Unfortunately according to the website my camera doesn’t support binning. I’m using a a2a1920-51gcpro gige camera.

For anyone interested, there’s another feature called Pixel Beyond that reduces the frame resolution on ace 2 cameras.

Hi, I'm getting a similar error at this line: `divisor = torch.tensor(list(range(1, input_shape[2] + 1))*input_shape[1], device='cuda').view([1, input_shape[1], input_shape[2], 1, 1])` I can't export to onnx because apparently adaptive pooling is...

@herry123435 Unfortunately no! I still have this issue, I ended up changing the whole network for now.

UPDATE: I noticed a very strange behavior. It seems like that torch-tensorrt is in fact speeding the model up a lot (about 3.5 times). However, generating the random tensor in...

@narendasan Thank you for your response. So far it seems like what causing the issue is the Modulus operator (%). The following network gives the same error: ``` class SomeNet(nn.Module):...

@narendasan Sorry for the late reply! Yes, I'm getting the same graph but tensorrt still gives the same error. ``` graph(%self : __torch__.MoViNet_pytorch.movinets.models.SomeNet, %x.1 : Tensor): %16 : bool =...