MMLabExample

About OpenMMLab

• official web site
• github
• platform
• twitter
• zhihu
• CVPR 2021 Tutorial

Download Dataset

COCO can be downloaded from COCO homepage. Also mmdet has a download script.

CUB can be downloaded from CUB homepage.

coco
├── annotations
├── train2017
└── val2017

cub
└── CUB_200_2011
    ├── image_class_labels.txt
    ├── images
    ├── images.txt
    └── train_test_split.txt

Environment setup

export CUB_DIR="/path/to/cub"
export COCO_DIR="/path/to/coco"

How to Run

Cifar10

docker-compose up -d mmcls_example

# Train
docker-compose exec mmcls_example mim train mmcls configs/cls/cifar/resnet18_1xb16_cifar10.py --gpus 1
# Test
docker-compose exec mmcls_example mim test mmcls configs/cls/cifar/resnet18_1xb16_cifar10.py --gpus 1 --checkpoint work_dirs/resnet18_1xb16_cifar10/epoch_200.pth --metrics accuracy

docker-compose stop mmcls_example
docker-compose rm mmcls_example

CUB

docker-compose up -d mmcls_example

# Train
docker-compose exec mmcls_example mim train mmcls configs/cls/cub/resnet50.py --gpus 1
# Test
docker-compose exec mmcls_example mim test mmcls configs/cls/cub/resnet50.py --gpus 1 --checkpoint work_dirs/resnet50/epoch_100.pth --metrics accuracy

docker-compose stop mmcls_example
docker-compose rm mmcls_example

COCO

docker-compose up -d mmdet_example

# Train
docker-compose exec mmdet_example mim train mmdet configs/det/faster_rcnn_r50_fpn_1x_coco.py --gpus 2 --launcher pytorch
# Test
docker-compose exec mmdet_example mim test mmdet configs/det/faster_rcnn_r50_fpn_1x_coco.py --gpus 1 --checkpoint work_dirs/faster_rcnn_r50_fpn_1x_coco/epoch_12.pth --eval bbox

docker-compose stop mmdet_example
docker-compose rm mmdet_example