CornerNet

Model description

CornerNet, a new approach to object detection where we detect an object bounding box as a pair of keypoints, the top-left corner and the bottom-right corner, using a single convolution neural network. By detecting objects as paired keypoints, we eliminate the need for designing a set of anchor boxes commonly used in prior single-stage detectors. In addition to our novel formulation, we introduce corner pooling, a new type of pooling layer that helps the network better localize corners. Experiments show that CornerNet achieves a 42.2% AP on MS COCO, outperforming all existing one-stage detectors.

Step 1: Installation

CornerNet model is using MMDetection toolbox. Before you run this model, you need to setup MMDetection first.

# Go to "toolbox/MMDetection" directory in root path
cd ../../../../toolbox/MMDetection/
bash install_toolbox_mmdetection.sh

Step 2: Preparing datasets

Go to visit COCO official website, then select the COCO dataset you want to download.

Take coco2017 dataset as an example, specify /path/to/coco2017 to your COCO path in later training process, the unzipped dataset path structure sholud look like:

coco2017
├── annotations
│   ├── instances_train2017.json
│   ├── instances_val2017.json
│   └── ...
├── train2017
│   ├── 000000000009.jpg
│   ├── 000000000025.jpg
│   └── ...
├── val2017
│   ├── 000000000139.jpg
│   ├── 000000000285.jpg
│   └── ...
├── train2017.txt
├── val2017.txt
└── ...

Step 3: Training

# Make soft link to dataset
cd mmdetection/
mkdir -p data/
ln -s /path/to/coco2017 data/coco

# On single GPU
python3 tools/train.py configs/cornernet/cornernet_hourglass104_mstest_8x6_210e_coco.py

# Multiple GPUs on one machine
bash tools/dist_train.sh configs/cornernet/cornernet_hourglass104_mstest_8x6_210e_coco.py 8

Results

Reference

• Cornernet: Detecting objects as paired keypoints