add bubdle input into AIBench (#64557)
authorWenliang Zhao <wenliangzhao@fb.com>
Tue, 7 Sep 2021 21:09:31 +0000 (14:09 -0700)
committerFacebook GitHub Bot <facebook-github-bot@users.noreply.github.com>
Tue, 7 Sep 2021 21:46:23 +0000 (14:46 -0700)
commit3bd69d30204df7d5d2f1254247c3e917aca8ce93
tree4a80834a546f82808315cb2048d111696ea21d5e
parent3c87f557523128a524f71168a21942d022e87883
add bubdle input into AIBench (#64557)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/64557

MaskRCNN speed depends on how many people detected in the detection stage. A random input from dataloader doesn't satisfy this. In order to standardize the benchmarking, we use 2 standard image for benchmarking, 2/3 people.

Test Plan: AIBench result: https://www.internalfb.com/intern/aibench/details/945883114818980

Reviewed By: axitkhurana

Differential Revision: D30446049

fbshipit-source-id: a2826fdb69e9f840c0afc566c4cbbcde1c2fba89
torch/utils/bundled_inputs.py