Misha Tran

Swin-Transformer mmdetection tensorrt

@VVsssssk do you know when swin-transformers can release

I check the tensorrt wrapper and output of this function always have 1 sample output. Plz check the forward method of it

input_img shape is `6*3*800*1088` ``` input_img = data["img"][0].contiguous() img_metas = data["img_metas"] outputs = self.model.forward_test( input_img, img_metas, return_loss=False, rescale=True) ``` But the output is only 1 images process

i have 1 product but many production date So attributes not a perfect solution i think

Nice idea! Which topic triton stored in and how to use it. I think you should add it on README file

What's the main purpose to use kafka in this case ?

Did we got anything there? Kafka is nice idea for multi triton server

it cannot pass to server. I got this error on client: ``` tritonclient.utils.InferenceServerException: [StatusCode.INTERNAL] Unable to open shared memory region: '/tVHIApxKugk6TwsknLr1b' ```

i run client in host and triton on docker. I set `--ipc host` on triton