search

mlcommons / training_policies

Issues related to MLPerf™ training policies, including rules and suggested changes
https://mlcommons.org/en/groups/training
Apache License 2.0
92 stars 66 forks source link

Mask-RCNN clarifiation around num_image_candidates #331

Open jonathan-cohen-nvidia opened 4 years ago

jonathan-cohen-nvidia commented 4 years ago

There was a rule previously that you could choose to do either:

Process the whole batch at a time and generate the total required number of patches (1000 * batches per chip)

OR

one image at a time, with exactly 1000 patches per image

Current HP Table (https://docs.google.com/spreadsheets/d/128r_G1KEE0C1PzuTnBGybeNsT-mXeORZhmWnqDgu8Ok/edit#gid=540034374 row 30 in the "Approved HP" tab) omits this detail. We should probably codify it somewhere.

@bitfort

bitfort commented 4 years ago

SWG:

Will follow up over email. Who wants 2000?

bitfort commented 4 years ago

AI(Victor) Email

bitfort commented 4 years ago

SWG:

This has been resolved in the HP table. Do what the table says.