DETR-Object detection配置教程

DETR-Object detection配置教程

0. github下载为zip压缩包，解压到你的工作目录下

Object Detection with DETR : A minimal implementation DEMO点击此处

1. Anaconda环境配置

• 创建detr环境

conda create -n detr python=3.8
conda activate detr
cd D:\Coding\AnacondaProject\detr-main #进入detr-main目录

• 安装pytorch 1.5+ 和torchvision 0.6+

conda install -c pytorch pytorch torchvision

• 安装pycocotools和scipy，用于在COCO数据集上进行evaluation及训练

conda install cython scipy
pip install -U 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI' #本机测试不行
pip install pycocotools #使用这条命令安装pycocotools

• （可选）如果想使用panoptic

pip install git+https://github.com/cocodataset/panopticapi.git

2. 数据准备

Download and extract COCO 2017 train and val images with annotations from http://cocodataset.org. We expect the directory structure to be the following:

path/to/coco/
  annotations/  # annotation json files
  train2017/    # train images
  val2017/      # val images

3. 训练

在单节点8个gpu上对基线DETR训练300个epochs：

python -m torch.distributed.launch --nproc_per_node=8 --use_env main.py --coco_path /path/to/coco

A single epoch takes 28 minutes, so 300 epoch training takes around 6 days on a single machine with 8 V100 cards. To ease reproduction of our results we provide results and training logs for 150 epoch schedule (3 days on a single machine), achieving 39.5/60.3 AP/AP50.

4. 评估

在COCO val 5k用一个GPU对DETR R50进行评估：

python main.py --batch_size 2 --no_aux_loss --eval --resume https://dl.fbaipublicfiles.com/detr/detr-r50-e632da11.pth --coco_path /path/to/coco

5. 多节点训练

分布式训练可以使用Slurm和submitit

pip install submitit

在4个节点上对基线模型DETR-6-6训练300 epochs：

python run_with_submitit.py --timeout 3000 --coco_path /path/to/coco