Skip to content

The SFPN is a novel plug-and-play component for the CNN object detector. This project is the official code for the paper "SFPN: Synthetic FPN for Object Detection" in IEEE ICIP 2022.

License

Notifications You must be signed in to change notification settings

D0352276/SFPN-Synthetic-FPN-for-Object-Detection

Repository files navigation

SFPN: Synthetic FPN for Object Detection

license

This project was developed from CSL-YOLO and is its follow-up study. The proposed SFPN (Synthetic Fusion Pyramid Network) arichtecture creates various synthetic layers between layers of the original FPN to enhance the accuracy of light-weight CNN backones to extract objects' visual features more accurately.

This paper has been accepted by IEEE ICIP 2022.

Paper Link: https://arxiv.org/abs/2203.02445

Architecture of Synthetic Fusion Module (SFM). The blue, orange and gray arrows represent the optional inputs. The experiments prove the SFPN architecture outperforms either the large backbone VGG16, ResNet50 or light-weight backbones such as MobilenetV2 based on AP score.

Synthetic Fusion Pyramid Network with synthetic outputs layers: (a) SFPN-5-SOL, (b) SFPN-9-SOL. The yellow squares represent SFBs.

More Suitable Feature

In this example, we found that the synthetic layers fit some objects better than original layers. The top three frames are the confidence maps output by SFPN-3, and the bottom three frames are the confidence maps output by SFPN-5-SOL.

Requirements

How to Reproduce the Results ?

#Train
python3 main.py -t cfg/train_coco.cfg

#MS-COCO Evaluation
python3 main.py -ce cfg/m2_224_3.cfg

#FPS Evaluation
python3 main.py -fe cfg/m2_224_3.cfg

#Drawing Confidence Figure
python3 main.py -pc cfg/m2_224_3.cfg

The cfg file can be changed by the paper prosoed versions. It has options like "m2_224_3.cfg", "m2_320_5.cfg", ......, and "vgg_320_9.cfg".

Fully Model Weights Files

Download Link

This link contains all weights of CNN models mentioned in the paper.

Fully Dataset

The entire MS-COCO dataset is too large, here only a few pictures are stored for DEMO,

if you need complete data, please download on this page.

Results

Comparison of SFPN-5 and SFPN-9 with the baseline (SFPN-3) on MS-COCO. Note that the FPS is derived from the I7-6600U.

About

The SFPN is a novel plug-and-play component for the CNN object detector. This project is the official code for the paper "SFPN: Synthetic FPN for Object Detection" in IEEE ICIP 2022.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages