Exact Recipe for Training on Coco from scratch

[subhankar-ghosh]

Can anyone please provide the exact settings (hyperparameters, schedulers, optimizer etc) for achieving the SOTA mAP with any of the architectures (D0, D1 etc any of these). With D0 90k training steps Global batch size 16 Number of GPUs 8

I am getting loss: 0.315 mAP = 2.716 small mAP = 0.8 med mAP = 2.59 large mAP = 4.45

[factplay1]

So I concluded that the code, "as is", does not reproduce COCO training from scratch (efficientdet-0).

Here is what I tried:

• Around 220+ training jobs with different configurations
• hyper-parameter tuning
• 1,2,4,8 GPU cases with various ranges of batch size
• Tried making some changes in the code so it can load pretrained efficientnet-d0 weights. This improved the loss, but still not low enough - lowest it got is 0.305, and even that gives low mAP (4%)
• Tried different kinds of optimizers, different kinds of LR scheduling techniques
• Tried 2-step training, where you train with adamw first, then try to further finetune with sgd
• Tried with the latest code as well as throughout the past 3 weeks or so
• Tried random seed vs fixed seed
• Tried different kinds of pretrained weights

I concluded that for custom datasets it works, but not for COCO training from scratch for research purposes.

[claire-s11]

Hi @subhankar-ghosh , by mAP = 2.716, do you mean iou= 0.5:0.95, map = 0.2716? Thank you!