Deep Patel

@djairhogeuens If you don't mind, can you please share the workaround. is it to restart the broker thread and reconnect?

I just noticed there is listener function for receiver loop as well which could be used to reconnect the broker. Haven't tested with it yet though. ```py def on_receiver_loop_completed(self, frame):...

One reason could be Computer/GPU algorithms in general are optimized for values in powers of 2 because that way we can utilize the memory access in the most efficient manner.

@qiuqiangkong We have faced similar issue trying to create a baidu account outside of China. It will be great if the dataset can be uploaded to some other platform. Thank...

> Any updates in supporting `List[Tensor]` as output? My usecase is object detection for which you may have different number of objects as outputs say image 1 has 2 objects...

Noticed similar issue on personal machine with `pytorch:1.9 and torchvision:0.10.0`