Science Score: 44.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • CITATION.cff file
    Found CITATION.cff file
  • codemeta.json file
    Found codemeta.json file
  • .zenodo.json file
    Found .zenodo.json file
  • DOI references
  • Academic publication links
  • Academic email domains
  • Institutional organization owner
  • JOSS paper metadata
  • Scientific vocabulary similarity
    Low similarity (10.8%) to scientific vocabulary
Last synced: 10 months ago · JSON representation ·

Repository

Basic Info
  • Host: GitHub
  • Owner: aitor-martinez-seras
  • License: agpl-3.0
  • Language: Python
  • Default Branch: main
  • Size: 27.6 MB
Statistics
  • Stars: 0
  • Watchers: 1
  • Forks: 0
  • Open Issues: 0
  • Releases: 0
Created over 1 year ago · Last pushed about 1 year ago
Metadata Files
Readme Contributing License Citation Security

README.md

Introduction

This is the code for the paper On the Inherent Robustness of One-Stage Object Detection against Out-of-Distribution Data.

The code is based on an early version of the Ultralytics YOLOv8 repository. The more updated version of the libray can be found here.

The version of ultralytics is 8.3.127.

Instalation of the package

To install the package:

  1. Install torch>=1.8 and torchvision>=0.8.1, with CUDA support preferably
  2. pip install -r requirements.txt
  3. If you want to use CLI commands, use pip install -e .

Basic structure

bash ├── datasets │ └──coco └── < source_code_folder > ├── datasets_utils ├── data_utils.py ├── examples ├── ultralytics ├── ultralytics.egg-info ├── venv-yolo ├── visualization_utils.py ... └── yolov8n.pt

Datasets must be outside the source code folder

Warnings

The code has been only tested on Linux and some of the code may not work, as some parts are not OS agnostic due to the usage of the / for splitting paths.

Download datasets

OWOD (Open World Object Detection)

Follow instructions in datasets_utils/owod/instructions.md.

COCO OOD and COCO Mixed

Follow instructions in datasetsutils/cocoood/instructions.md.

Usage

The model obtaining the results of the paper can be downloaded from this OneDrive link.

Replicate results

1) Train a model using the replicate/train.sh. The arguments are the same to the ones used for the paper.

2) Run the bash scripts inside the replicate/benchmarks.sh to obtain the results.

3) To obtain the figures of the paper, run the process_results.ipynb. The code expects the following structure:

```bash
< source_code_folder >
├── results
    ├── fmap_method
    │   └── < .csv files from the vanilla FMap method >
    ├── fmap_method_SDR
    │   └── < .csv files from the SDR FMap method >
    ├── fmap_method_EUL
    │   └── < .csv files from the EUL FMap method >
    ├── logits_methods
    │   └── < .csv files from the post-hoc or logits methods >
    └── fusion_methods
        └── < .csv files from the fusion methods >
```

Using the OOD methods

Examples of usage are provided in the scripts/ood_evaluation.sh file.

Training a YOLO model

Examples are provided scripts/train_owod.sh file.

Owner

  • Name: Aitor Martínez Seras
  • Login: aitor-martinez-seras
  • Kind: user
  • Location: Bilbao
  • Company: University of Deusto

PhD candidate at TECNALIA Research & Innovation

Citation (CITATION.cff)

cff-version: 1.2.0
preferred-citation:
  type: software
  message: If you use this software, please cite it as below.
  authors:
  - family-names: Jocher
    given-names: Glenn
    orcid: "https://orcid.org/0000-0001-5950-6979"
  - family-names: Chaurasia
    given-names: Ayush
    orcid: "https://orcid.org/0000-0002-7603-6750"
  - family-names: Qiu
    given-names: Jing
    orcid: "https://orcid.org/0000-0003-3783-7069"
  title: "YOLO by Ultralytics"
  version: 8.0.0
  # doi: 10.5281/zenodo.3908559  # TODO
  date-released: 2023-1-10
  license: AGPL-3.0
  url: "https://github.com/ultralytics/ultralytics"

GitHub Events

Total
  • Push event: 3
  • Create event: 1
Last Year
  • Push event: 3
  • Create event: 1