This repository is the official implementation of paper Dynamic Curriculum Learning for Great Ape Detection in the Wild.
-
Linux, CUDA>=9.2, GCC>=5.4
-
Python>=3.7
- create a conda environment:
conda create -n DCLNet python=3.7 pip- activate the environment:
conda activate DCLNet- Installation pytorch follow the official doc install at here
conda install pytorch==1.8.0 torchvision==0.9.0 cudatoolkit=10.2 -c pytorch- Installation other packages
pip install -r requirements.txt- Install CUDA extension
cd ./models/ops
sh ./make.sh- test your CUDA extension
python test.py
CPU: AMD EPYC 7543 (128) @ 2.794GHz
GPU: NVIDIA GRID A100X
OS: CentOS Linux release 7.9.2009 (Core) x86_64
nvcc version: 11.2
GCC version: 7.5.0
python version: 3.7.11
Please download COCO 2017 dataset and organize it in the following structure:
data/
└── coco
├── test2017
├── train2017
├── val2017
└── annotations
├── instances_train2017.json
├── instances_val2017.json
└── semi_supervised
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
├── [email protected]
└── [email protected]
The partial labelled data (PLD) are split by:
cd ./datasets/prepare_coco
sh ./run.shComing soon! Stay tuned !
- running with 1% PLD at fold1 (4 high-end GPUs required):
sh ./train_coco_1percent_1fold.sh- running with 5% PLD at fold1 (4 high-end GPUs required):
sh ./train_coco_5percent_1fold.sh- running with 10% PLD at fold1 (4 high-end GPUs required):
sh ./train_coco_10percent_1fold.shComing soon! Stay tuned
| Setting | Dataset | mAP | mAP50 | map75 | checkpoint (Student and Teacher) |
|---|---|---|---|---|---|
| 1% PLD | COCO | 17.34 | 31.00 | 17.35 | ckpt |
| 5% PLD | COCO | 29.75 | 46.68 | 31.77 | ckpt |
| 10% PLD | COCO | 34.45 | 51.93 | 37.14 | ckpt |
| 10% PLD | PanAfrican | 45.96 | 78.10 | 47.67 | coming soon |
| 20% PLD | PanAfrican | 59.01 | 89.23 | 66.95 | coming soon |
| 50% PLD | PanAfrican | 63.39 | 92.96 | 70.00 | coming soon |
If you are considering using this codebase, please cite our work:
@misc{yang2021dcldet,
title={Dynamic Curriculum Learning for Great Ape Detection in the Wild},
author={Xinyu Yang and Tilo Burghardt and Majid Mirmehdi},
year={2022},
eprint={2205.00275},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
This repository builds on previous works codebase 1. Deformable DETR 2. DETReg. Please consider citing these works as well.
