data

Data & Pre-trained Models

Put the datasets and the pre-trained models here.

1. Prepare datasets

• PASCAL-5i
• COCO-20i

1.1 PASCAL-5i

• Download Training/Validation data (2G), and extract VOCtrainval_11-May-2012.tar to ./data/
• Download CaNet annotations (20.7M), and extract Binary_map_aug.zip to ./data/VOCdevkit/VOC2012/

1.2 COCO-20i

• Download 2014 Training images (13GB), 2014 Val images (6GB), 2014 Train/Val annotations (241M), and extract them to ./data/COCO

2. Pre-trained weights

• VGG-16

• Resnet-50

• Resnet-101

• Resnet50_v2 (for PFENet. See the PFENet repository for details.)

• Download pre-trained models and put them into ./data .

Final directory structure (only display used directories and files):

./data
├── COCO
│   ├── annotations
│   ├── train2014
│   └── val2014
├── VOCdevkit
│   └── VOC2012
│       ├── Binary_map_aug
│       └── JPEGImages
├── README.md
├── resnet101-5d3b4d8f.pth
├── resnet50-19c8e357.pth
├── resnet50_v2.pth
└── vgg16-397923af.pth