EfficientDet-Pytorch-docker

A docker build file for EfficientDet: https://github.com/zylo117/Yet-Another-EfficientDet-Pytorch.git

Requirements

Nvidia Docker runtime: https://github.com/NVIDIA/nvidia-docker#quickstart
CUDA 10.1 or higher on your host, check with nvidia-smi

Prepare your dataset

your dataset structure should be like this

datasets/
    -your_project_name/
        -train_set_name/
            -*.jpg
        -val_set_name/
            -*.jpg
        -annotations
            -instances_{train_set_name}.json
            -instances_{val_set_name}.json

for example, coco2017

datasets/
    -coco2017/
        -train2017/
            -000000000001.jpg
            -000000000002.jpg
            -000000000003.jpg
        -val2017/
            -000000000004.jpg
            -000000000005.jpg
            -000000000006.jpg
        -annotations
            -instances_train2017.json
            -instances_val2017.json

Host machine

Build image
```
docker build . -t efficientdet:latest
```

Run container

docker run -it --rm --runtime=nvidia efficientdet:latest /bin/bash

  export WORK_DIR=~/Documents/efficientDet/ && \
  mkdir -p $WORK_DIR/weights && \
  mkdir -p $WORK_DIR/logs && \
  docker run  -v $WORK_DIR/weights:/efficientdet/weights \
  -v $WORK_DIR/datasets:/datasets \
  -v $WORK_DIR/logs:/logs \
  -v /etc/localtime:/etc/localtime \
  -v /tmp/.X11-unix:/tmp/.X11-unix \
  -e DISPLAY=$DISPLAY -e QT_X11_NO_MITSHM=1  \
  -it --rm --runtime=nvidia \
  --name efficientdet efficientdet:latest  /bin/bash

Download pretrain weights

  export WEIGHT_DIR=~/Documents/efficientDet/weights/
  wget -nc -P $WEIGHT_DIR https://github.com/zylo117/Yet-Another-Efficient-Pytorch/releases/download/1.0/efficientdet-d0.pth
  wget -nc -P $WEIGHT_DIR https://github.com/zylo117/Yet-Another-Efficient-Pytorch/releases/download/1.0/efficientdet-d1.pth
  wget -nc -P $WEIGHT_DIR https://github.com/zylo117/Yet-Another-Efficient-Pytorch/releases/download/1.0/efficientdet-d2.pth
  wget -nc -P $WEIGHT_DIR https://github.com/zylo117/Yet-Another-Efficient-Pytorch/releases/download/1.0/efficientdet-d3.pth
  wget -nc -P $WEIGHT_DIR https://github.com/zylo117/Yet-Another-Efficient-Pytorch/releases/download/1.0/efficientdet-d4.pth
  wget -nc -P $WEIGHT_DIR https://github.com/zylo117/Yet-Another-Efficient-Pytorch/releases/download/1.0/efficientdet-d5.pth
  wget -nc -P $WEIGHT_DIR https://github.com/zylo117/Yet-Another-Efficient-Pytorch/releases/download/1.0/efficientdet-d6.pth

Converting a training model to inference model
- Running directly from the repository:
  
  keras_retinanet/bin/convert_model.py /path/to/training/model.h5 /path/to/save/inference/model.h5
- Using the installed script:
  
  retinanet-convert-model /path/to/training/model.h5 /path/to/save/inference/model.h5

Save and Load docker image

  docker save retinanet > retinanet.tar
  docker load < retinanet.tar

In container

Train Coco dataset in container

  cd /efficientdet && \
  python3 train.py --project item \
    -c 0 \
    --data_path /datasets/ \
    --num_workers 2 \
    --batch_size 2 \
    --log_path /logs \
    --saved_path /logs \
    --num_epochs 500

Resume training from a snapshot

  cd /retinanet/keras_retinanet/bin && \
  python3 train.py --snapshot /snapshots/resnet50_coco_50.h5 \
    --initial-epoch 50 --epochs 100 \
    coco /datasets/coco/

The retinaNet repo is in /retinanet

Possible issue

If meet the following error : cannot connect to X server :1