sparsercnn

DGMN2 with Sparse R-CNN

This folder contains the implementation of DGMN2 for object detection with Sparse R-CNN.

Results

COCO validation set

Method	Backbone	Lr schd	AP	Config	Download
Sparse R-CNN	DGMN2-Small	3x	48.2	config	model

Getting Started

Clone the repository locally:

git clone https://github.com/fudan-zvg/DGMN2

Installation

a. Install Detectron2 following the official instructions. Here we use Detectron2 0.4.

b. Install PyTorch Image Models. Here we use PyTorch Image Models 0.4.5.

pip install timm==0.4.5

c. Build the extension.

cd dcn
python setup.py build_ext --inplace

Data preparation

First, prepare COCO dataset according to the guidelines in Detectron2.

Then, download the weights pretrained on ImageNet, and put them in a folder pretrained/.

Training

To train DGMN2-Small + Sparse R-CNN using 300 learnable proposals on COCO train2017 on a single node with 8 GPUs for 36 epochs run:

python train_net.py --num-gpus 8 --config-file configs/sparsercnn.dgmn2small.300pro.3x.yaml

Evaluation

To evaluate DGMN2-Tiny + RetinaNet on COCO val2017 on a single node with 8 GPUs run:

python train_net.py --num-gpus 8 --config-file --config-file configs/sparsercnn.dgmn2small.300pro.3x.yaml --eval-only MODEL.WEIGHTS path/to/checkpoint_file