efficientdet-pytorch
A Pytorch implementation of EfficientDet.
Pytorch implementtation of EfficientDet object detection as described in EfficientDet: Scalable and Efficient Object Detection
This implementation is a very simple version without many data augmentation.
The EfficientNet code are borrowed from the A PyTorch implementation of EfficientNet,if you want to train EffcicientDet from scratch,you should load the efficientnet pretrained parameter. use
python train.py --coco_path '/home/hoo/Dataset/COCO' --backbon 'efficientnet-b0' --backbone_pretrained True
and the efficientnet pretrainied parameter will be download and load automatically, and start to train.
I've only trained efficientdet-d0 so far,and without many data augmentation.if you want to load efficientnet pretrained parameter,use
python train.py --coco_path '/home/hoo/Dataset/COCO' --backbone 'efficientnet-b0' --backbone_pretrained False --EfficientDet_pretrained True --pretrained './weights/efficientdet_0.pth'
| Model | mAP | pre_trained |
|---|---|---|
| efficientdet-d0 | 23.2% | download |
