Skip to content

detrex v0.3.0 Release

Compare
Choose a tag to compare
@rentainhe rentainhe released this 17 Mar 09:47
· 45 commits to main since this release
69b50b8

What's New

New Algorithms

  • Support Anchor-DETR
  • Support DETA

More training techniques

  • Support EMAHook during training by setting train.model_ema.enabled=True, which can further enhance the model performance.
  • Fully support mixed precision training by setting train.amp.enabled=True, which can reduce 20% to 30% GPU memory usage.
  • Support encoder-decoder checkpoint in DINO which may reduce 30% GPU memory usage. And for more details about the checkpoint usage, please refer to this PR: #200
  • Support fast debugging by setting train.fast_dev_run=True.
  • Support a great slurm training scripts by @rayleizhu , please check this issue #213 for more details.

Release more than 10+ pretrained checkpoints

Method Pretrained Epochs Box AP
DETR-R50-DC5 IN1k 500 43.4
DETR-R101-DC5 IN1k 500 44.9
Anchor-DETR-R50 IN1k 50 42.2
Anchor-DETR-R50-DC5 IN1k 50 44.2
Anchor-DETR-R101 IN1k 50 43.5
Anchor-DETR-R101-DC5 IN1k 50 45.1
Conditional-DETR-R50-DC5 IN1k 50 43.8
Conditional-DETR-R101 IN1k 50 43.0
Conditional-DETR-R101 -DC5 IN1k 50 45.1
DAB-DETR-R50-3patterns IN1k 50 42.8
DAB-DETR-R50-DC5 IN1k 50 44.6
DAB-DETR-R50-DC5-3patterns IN1k 50 45.7
DAB-DETR-101-DC5 IN1k 50 45.7
DN-DETR-R50-DC5 IN1k 50 46.3
DINO with EMA IN1k 12 49.4
DETA-R50-5scale IN1k 12 50.1
DETA-Swin-Large object-365 24 62.9

Part of the pre-trained weights are converted from their official repo, and all the pre-trained weights can be downloaded in detrex Model Zoo