Skip to content

Latest commit

 

History

History
169 lines (136 loc) · 7.08 KB

README.md

File metadata and controls

169 lines (136 loc) · 7.08 KB
Raw

Dynamic Curriculum Learning for Great Ape Detection in the Wild

Xinyu Yang, Tilo Burghardt, Majid Mirmehdi

DCLDET

This repository is the official implementation of paper Dynamic Curriculum Learning for Great Ape Detection in the Wild.

Installation

Requirements

  • Linux, CUDA>=9.2, GCC>=5.4

  • Python>=3.7

Install Dependencies

  1. create a conda environment:
conda create -n DCLNet python=3.7 pip
  1. activate the environment:
conda activate DCLNet
  1. Installation pytorch follow the official doc install at here
conda install pytorch==1.8.0 torchvision==0.9.0 cudatoolkit=10.2 -c pytorch
  1. Installation other packages
pip install -r requirements.txt
  1. Install CUDA extension
cd ./models/ops
sh ./make.sh
  1. test your CUDA extension
python test.py

Test Environments

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

Usage

Dataset preparation

MSCoco

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
            ├── instances_train2017.1@10.json
            ├── instances_train2017.1@10-unlabeled.json
            ├── instances_train2017.1@1.json
            ├── instances_train2017.1@1-unlabeled.json
            ├── instances_train2017.1@5.json
            ├── instances_train2017.1@5-unlabeled.json
            ├── instances_train2017.2@10.json
            ├── instances_train2017.2@10-unlabeled.json
            ├── instances_train2017.2@1.json
            ├── instances_train2017.2@1-unlabeled.json
            ├── instances_train2017.2@5.json
            ├── instances_train2017.2@5-unlabeled.json
            ├── instances_train2017.3@10.json
            ├── instances_train2017.3@10-unlabeled.json
            ├── instances_train2017.3@1.json
            ├── instances_train2017.3@1-unlabeled.json
            ├── instances_train2017.3@5.json
            ├── instances_train2017.3@5-unlabeled.json
            ├── instances_train2017.4@10.json
            ├── instances_train2017.4@10-unlabeled.json
            ├── instances_train2017.4@1.json
            ├── instances_train2017.4@1-unlabeled.json
            ├── instances_train2017.4@5.json
            ├── instances_train2017.4@5-unlabeled.json
            ├── instances_train2017.5@10.json
            ├── instances_train2017.5@10-unlabeled.json
            ├── instances_train2017.5@1.json
            ├── instances_train2017.5@1-unlabeled.json
            ├── instances_train2017.5@5.json
            └── instances_train2017.5@5-unlabeled.json

The partial labelled data (PLD) are split by:

cd ./datasets/prepare_coco
sh ./run.sh

PanAfrican

Coming soon! Stay tuned !

Semi-supervised Mix-training

COCO training

  1. running with 1% PLD at fold1 (4 high-end GPUs required):
sh ./train_coco_1percent_1fold.sh
  1. running with 5% PLD at fold1 (4 high-end GPUs required):
sh ./train_coco_5percent_1fold.sh
  1. running with 10% PLD at fold1 (4 high-end GPUs required):
sh ./train_coco_10percent_1fold.sh

PanAfrican training

Coming soon! Stay tuned

Models checkpoints

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

Citation

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}
}

Credits

This repository builds on previous works codebase 1. Deformable DETR 2. DETReg. Please consider citing these works as well.