/dpp

Primary LanguagePythonMIT LicenseMIT

Should All Proposals be Treated Equally in Object Detection? (ECCV 2022)

A pytorch implementation of DPP.

Requirements

  • Linux or macOS with Python ≥ 3.7
  • PyTorch = 1.12 and torchvision that matches the PyTorch installation.

Models

Model AP Head FLOPS download
DPP-XL 45.0 15G init / model
DPP-L 43.7 6.8G init / model
DPP-M 42.2 3.2G init / model
DPP-S 40.4 2.1G init / model

Steps

  1. Install and build libs
git clone https://github.com/liyunsheng13/dpp.git
cd dpp
python setup.py build develop
  1. Link coco dataset path to dpp/datasets/coco
mkdir -p datasets/coco
ln -s /path_to_coco_dataset/annotations datasets/coco/annotations
ln -s /path_to_coco_dataset/train2017 datasets/coco/train2017
ln -s /path_to_coco_dataset/val2017 datasets/coco/val2017
  1. Train DPP
python projects/DPP/train_net.py --num-gpus 8 \
    --config-file projects/DPP/configs/dpp.l.resnet.3x.yaml MODEL.WEIGHTS /path/to/init_model.pth
  1. Evaluate DPP
python projects/DPP/train_net.py --num-gpus 8 \
    --config-file projects/DPP/configs/dpp.l.resnet.3x.yaml \
    --eval-only MODEL.WEIGHTS /path/to/model.pth