EfficientDet: Scalable and Efficient Object Detection, in PyTorch

A PyTorch implementation of EfficientDet from the 2019 paper by Mingxing Tan Ruoming Pang Quoc V. Le Google Research, Brain Team. The official and original: comming soon.

Fun with Demo:

python demo.py --weight ./checkpoint_VOC_efficientdet-d1_97.pth --threshold 0.6 --iou_threshold 0.5 --cam --score

Table of Contents

• Recent Update
• Benchmarking
• Installation
• Installation
• Prerequisites
• Datasets
• Train
• Evaluate
• Performance
• Demo
• Future Work
• Reference

       

Recent Update

• [7/12/2019] Support Efficient-D0, Efficient-D1, Efficient-D2, Efficient-D3, Efficient-D4,... . Support change gradient accumulation steps, AdamW.

Benchmarking

We benchmark our code thoroughly on three datasets: pascal voc and coco, using family efficientnet different network architectures: EfficientDet-D0->7. Below are the results:

1). PASCAL VOC 2007 (Train/Test: 07trainval/07test, scale=600, ROI Align)

model	#GPUs	batch size	lr	lr_decay	max_epoch	time/epoch	mem/GPU	mAP
EfficientDet-D1(with Weight)	2	32	1e-4	30	100	20.min	20100 MB	updating

Installation

• Install PyTorch by selecting your environment on the website and running the appropriate command.
• Clone this repository and install package prerequisites below.
• Then download the dataset by following the instructions below.
• Note: For training, we currently support VOC and COCO, and aim to add ImageNet support soon.

prerequisites

• Python 3.6+
• PyTorch 1.3+
• Torchvision 0.4.0+ (We need high version because Torchvision support nms now.)
• requirements.txt

Datasets

To make things easy, we provide bash scripts to handle the dataset downloads and setup for you. We also provide simple dataset loaders that inherit torch.utils.data.Dataset, making them fully compatible with the torchvision.datasets API.

VOC Dataset

PASCAL VOC: Visual Object Classes

Download VOC2007 + VOC2012 trainval & test

# specify a directory for dataset to be downloaded into, else default is ~/data/
sh datasets/scripts/VOC2007.sh
sh datasets/scripts/VOC2012.sh

COCO

Microsoft COCO: Common Objects in Context

Download COCO 2014

# specify a directory for dataset to be downloaded into, else default is ~/data/
sh datasets/scripts/COCO2014.sh

Note: Read dataset COCO will support soon.

Training EfficientDet

• To train EfficientDet using the train script simply specify the parameters listed in train.py as a flag or manually change them.

python train.py --model_name effcientdet-d0 # Example

• With VOC Dataset:

python train.py --dataset_root /root/data/VOCdevkit/ --model_name effcientdet-d0 # Example

• With COCO Dataset: Support soon

Evaluation

To evaluate a trained network:

python eval.py

Demo

python demo.py --weights ./checkpoint_VOC_efficientdet-d1_97.pth --threshold 0.5

Output:

Webcam Demo

You can use a webcam in a real-time demo by running:

python demo.py --weight ./checkpoint_VOC_efficientdet-d1_97.pth --threshold 0.6 --iou_threshold 0.5 --cam --score

Performance

TODO

We have accumulated the following to-do list, which we hope to complete in the near future

• Still to come:
  • EfficientDet-D0-7
  • GPU-Parallel
  • NMS
  • Soft-NMS
  • Pretrained model
  • Demo
  • Model zoo
  • TorchScript
  • Mobile
  • C++ Onnx

Authors

• Toan Dao Minh

Note: Unfortunately, this is just a hobby of ours and not a full-time job, so we'll do our best to keep things up to date, but no guarantees. That being said, thanks to everyone for your continued help and feedback as it is really appreciated. We will try to address everything as soon as possible.

References

• tanmingxing, rpang, qvl, et al. "EfficientDet: Scalable and Efficient Object Detection." EfficientDet.
• A list of other great EfficientDet ports that were sources of inspiration:
  • EfficientNet
  • SSD.Pytorch
  • RetinaNet.Pytorch
  • NMS.Torchvision

Citation

@article{efficientdetpytoan,
    Author = {Toan Dao Minh},
    Title = {A Pytorch Implementation of EfficientDet Object Detection},
    Journal = {github.com/toandaominh1997/EfficientDet.Pytorch},
    Year = {2019}
}