we offer a leaderborad for active learning in object detection in here
Active learning has been popular in image classification, but has not been fully explored in object detection. Most of current works on object detection are evaluated with different settings, making it difficult to fairly compare their performance. To facilitate the research in this field, this paper contributes an active learning benchmark framework named as ALBench for evaluating active learning in object detection. Developed on an automatic deep model training system, this ALBench framework is easy-to-use, compatible with different active learning algorithms, and ensures the same training and testing protocols. We hope this automated benchmark system help researchers to easily reproduce literature's performance and have fair comparisons with prior arts
├── download_dataset.py
├── infer-config.yaml
├── training-config.yaml
├── mining-config.yaml
├── labels.yaml
├── README.md
├── requirements.txt
├── run.py
├── run.sh
├── init_trainSet_5000.txt
├── command
├── deinit.sh
├── exclude.sh
├── import.sh
├── init.sh
├── join.sh
├── merge.sh
├── mining.sh
├── test.sh
└── training.sh
├── data
│ ├── COCO
│ │ ├──labels.yaml
│ │ ├── annotations
│ │ │ └── xml
│ │ ├── test2017
│ │ ├── train2017
│ │ └── val2017
│ ├── mining
│ │ ├── anno
│ │ └── img
│ ├── train
│ │ ├── anno
│ │ └── img
│ └── val
│ ├── anno
│ └── img
└── utils
└── generate_dataset.py
└── coco2voc.py
path/to/ALBecnh$ pip install -e ymir1.1.0/ymir/command
path/to/ALBecnh$ pip install -r requirements.txt
- Download COCO dataset
path/to/ALBecnh$ python utils/download_dataset.py
- Organize the dataset:
path/to/ALBecnh$ python utils/coco2voc.py
path/to/ALBecnh$ python utils/generate_dataset.py
dataset should be organized as following
ALBench/
├── data/
│ ├── COCO/
│ │ ├── mining/
│ │ │ ├── anno/
│ │ │ └── img/
│ │ ├── train/
│ │ │ ├── anno/
│ │ │ └── img/
│ │ ├── val/
│ │ │ ├── anno/
│ │ │ └── img/
ALBench is based on Ymir system, detector and active learning algorithm should be build as docker image. To build your own Ymir executor, or get our official executor, see build ymir executor for detail
- modify ALBench_config.yaml based on the introduction below
# your user name and password for ALBench Leaderboard if you want result automatically uploaded to leaderboard
user_name: ""
password: ''
# 1 if you using any other training strategy,such as semi-supervised learning , else 0
leaderboard_id: 1
# detector and training_docker are paired
detector : ['YOLOV5','SSD']
training_docker: ['youdaoyzbx/ymir-executor:ymir1.1.0-YOLOV5','youdaoyzbx/ymir-executor:ymir1.1.0-SSD']
# mining_algo and mining_docker are paired
mining_algo: [cald]
sh run.sh
training and mining result will be listed in
path/to/ALBecnh/tmp/training
path/to/ALBecnh/tmp/mining