~Please note this is only a beta release at this stage~
Fully convolutional one-stage object detection (FCOS) is a framework for per-pixel prediction of objects in images. FCOS doesn't rely on expensive anchor box calculations and their hyper-parameters, which is in contrast to state-of-the-art object detectors like RetinaNet, YOLOv3, and Faster R-CNN.
This repository contains an open-source implementation of FCOS in Python, with access to pre-trained weights for a number of different models. The package provides PyTorch implementations for using training, evaluation, and prediction in your own systems. The package is easily installable with conda, and can also be installed via pip if you'd prefer to manually manage dependencies.
Our code is free to use, and licensed under BSD-3. We simply ask that you cite our work if you use FCOS in your own research.
This repository brings the work from a number of sources together. Please see the links below for further details:
- our original paper: "FCOS: Fully convolutional one-stage object detection"
- our latest paper: "FCOS: A Simple and Strong Anchor-free Object Detector"
- the original FCOS implementation: https://github.com/tianzhi0549/FCOS
- implementation in the AdelaiDet toolbox: https://github.com/aim-uofa/AdelaiDet/blob/master/configs/FCOS-Detection/README.md#fcos-real-time-models
We offer three methods for installing FCOS:
- Through our Conda package: single command installs everything including system dependencies (recommended)
- Through our pip package: single command installs FCOS and Python dependences, you take care of system dependencies
- Directly from source: allows easy editing and extension of our code, but you take care of building and all dependencies
The only requirement is that you have Conda installed on your system, and NVIDIA drivers installed if you want CUDA acceleration. We provide Conda packages through Conda Forge, which recommends adding their channel globally with strict priority:
conda config --add channels conda-forge
conda config --set channel_priority strict
Once you have access to the conda-forge channel, FCOS is installed by running the following from inside a Conda environment:
conda install fcos
We don't explicitly lock the PyTorch installation to a CUDA-enabled version to maximise compatibility with our users' possible setups. If you wish to ensure a CUDA-enabled PyTorch is installed, please use the following installation line instead:
conda install pytorch=*=*cuda* fcos
You can see a list of our Conda dependencies in the FCOS feedstock's recipe.
Before installing via pip, you must have the following system dependencies installed if you want CUDA acceleration:
- NVIDIA drivers
- CUDA
Then FCOS, its custom CUDA code, and all of its Python dependencies, can be installed via:
pip install fcos
Installing from source is very similar to the pip method above, accept we install from a local copy. Simply clone the repository, enter the directory, and install via pip:
pip install -e .
Note: the editable mode flag (-e) is optional, but allows you to immediately use any changes you make to the code in your local Python ecosystem.
We also include scripts in the ./scripts directory to support running FCOS without any pip installation, but this workflow means you need to handle all system and Python dependencies manually.
FCOS can be used entirely from the command line, or through its Python API. Both call the same underlying implementation, and as such offer equivalent functionality. We provide both options to facilitate use across a wide range of applications. See below for details of each method.
When installed, either via pip or conda, a fcos executable is made available on your system PATH.
The fcos executable provides access to all functionality, including training, evaluation, and prediction. See the --help flags for details on what the command line utility can do, and how it can be configured:
fcos --help
fcos train --help
fcos evaluate --help
fcos predict --help
FCOS can also be used like any other Python package through its API. The API consists of a Fcos class with three main functions for training, evaluation, and prediction. Below are some examples to help get you started with FCOS:
from fcos import Fcos, fcos_config
# Initialise a FCOS network using the default 'FCOS_imprv_R_50_FPN_1x' model
f = Fcos()
# Initialise a FCOS network with the 'FCOS_imprv_dcnv2_X_101_64x4d_FPN_2x' model
f = Fcos(load_pretrained='FCOS_imprv_dcnv2_X_101_64x4d_FPN_2x')
# Create an untrained model with the settings for 'FCOS_imprv_R_101_FPN_2x'
f = Fcos(config_file=fcos_config('FCOS_imprv_R_101_FPN_2x'))
# Train a new model on the dataset specified by the config file (DATASETS.TRAIN)
f.train()
# Train a new model on a custom dataset, with a custom checkpoint frequency
f.train(dataset_name='custom_dataset', checkpoint_period=10)
# Get object detection boxes given an input NumPy image
detection_boxes = f.predict(image=my_image)
# Save an image with detection boxes overlaid to file, given an image file
f.predict(image_file='/my/detections.jpg',
output_file='/my/image.jpg')
# Evaluate your model's performance against the dataset specified by
# DATASETS.TEST in the config file, and output the results to a specific
# location
f.evaluate(output_directory='/my/eval/output/')If using FCOS in your work, please cite our original ICVV paper:
@inproceedings{tian2019fcos,
title={FCOS: Fully convolutional one-stage object detection},
author={Tian, Zhi and Shen, Chunhua and Chen, Hao and He, Tong},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
pages={9627--9636},
year={2019}
}Or our more recent TPAMI journal with further details of our work:
@article{tian2021fcos,
author={Tian, Zhi and Shen, Chunhua and Chen, Hao and He, Tong},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
title={FCOS: A Simple and Strong Anchor-free Object Detector},
year={2020},
volume={},
number={},
pages={1-1},
doi={10.1109/TPAMI.2020.3032166}}