Need to retrain the following models (with the best hyperparams selected by previous HPO runs) to recreate correct learning curves for the ARPA E review. Feel free to discuss model choices here and pick what seems best to you (we don't need to be too particular for the purposes of this presentation)
[ ] MaskRCNN IrO2 Finetune HPO (chosen because it had a slightly better mAP score and IOU, which is closer to what we truly want, even though it has worse loss)
[ ] FasterRCNN mAP HPO (feel free to pick MaskRCNN if that feels more appropriate)
Then, Slack Ari when done so she can put everything into beautiful slides :)
@jimtjames @NathanPruyne feel free to assign to whomever has the availability
ascourtas
commented
2 years ago
Need to retrain the following models (with the best hyperparams selected by previous HPO runs) to recreate correct learning curves for the ARPA E review. Feel free to discuss model choices here and pick what seems best to you (we don't need to be too particular for the purposes of this presentation)
Then, Slack Ari when done so she can put everything into beautiful slides :)
@jimtjames @NathanPruyne feel free to assign to whomever has the availability
Needed by EOD Friday 5/27