AutoShape behaviour is different for Tensor vs Image

[bmharper]

It took me some debugging to figure out that AutoShape has different behaviour when the input is a tensor vs a filename or a PIL Image. Specifically, when you feed AutoShape a tensor, then it skips the NMS and other post-processing, and just returns the raw tensor output.

I was wondering if it's worthwhile creating a PR which makes this behaviour optional, or perhaps if there is a place where we can document this?

Thanks

[glenn-jocher]

@bmharper yes good question. I think it makes sense to unify the behaviour. Can you please submit a PR for this?

[bmharper]

Great, I'll make a PR

[bmharper]

Looks like you've got a lot of PR workload!
Anyway.. I've added a PR for this: #10139
All ticks green. Tried my best not to break existing code/downstream users.

[glenn-jocher]

@bmharper got it, thank you!