Got the weird results which returned from Triton inference

AI & Data Science Deep Learning (Training & Inference) Triton Inference Server - archived
,
1

Dear all,

Recently I was trying to use the latest version of Triton server that it changed a lot of API functions.

Hence, I was followed the example from here: server/image_client.py at v2.8.0 · triton-inference-server/server · GitHub

I checked those examples which belong to classification, so when the model returns the results, there is only 1 value for each result. However, for my case, my model is object detection, so I expect that the results will be like a vector.

I got the results from the current triton server:

0.488281
0.707031
1.082031
0.144531
-3.988281
64.500000

Obviously, the results are scalars.

However, when I used the v20.03 which was an older version of TRTIS, the results were like below:

For example:

(,81111)
(,6458)
(,324)
(,159)
(,13135)
(,7358)

Is there any example or script about object detection? Or any suggestions?

Thank you in advance!!

BR,
Chieh

2

Resolved:

httpclient.InferRequestedOutput("output node name",binary_data=True)

Dont use class_count=classes