Skip to content
/ FSAF Public
forked from xuannianz/FSAF

FSAF (Feature Selective Anchor-Free Module for Single-Shot Object Detection) implementation in Keras and Tensorflow

Notifications You must be signed in to change notification settings

masati91/FSAF

 
 

Repository files navigation

FSAF

This is an implementation of FSAF on keras and Tensorflow. The project is based on fizyr/keras-retinanet and fsaf branch of zccstig/mmdetection. Thanks for their hard work.

As the authors write, FASF module can be plugged into any single-shot detectors with FPN-like structure smoothly. I have also tried on yolo3. Anchor-free yolo3(with FSAF) gets a comparable performance with the anchor-based counterpart. But you don't need to pre-compute the anchor sizes any more. And it is much better and faster than the one based on retinanet.

Updates

  • [03/05/2020] The author of the paper has released a new paper SAPD, which is based on FSAF. I have implemented it at xuannianz/SAPD.

Test

  1. I trained on Pascal VOC2012 trainval.txt + Pascal VOC2007 train.txt, and validated on Pascal VOC2007 val.txt. There are 14041 images for training and 2510 images for validation.
  2. The best evaluation results (score_threshold=0.05) on VOC2007 test are:
backbone mAP50
resnet50 0.7248
resnet101 0.7652
  1. Pretrained models are here.
    baidu netdisk extract code: rbrr
    goole dirver

  2. python3 inference.py to test your image by specifying image path and model path there.

image1 image2 image3

Train

build dataset (Pascal VOC, other types please refer to fizyr/keras-retinanet)

  • Download VOC2007 and VOC2012, copy all image files from VOC2007 to VOC2012.
  • Append VOC2007 train.txt to VOC2012 trainval.txt.
  • Overwrite VOC2012 val.txt by VOC2007 val.txt.

train

  • python3 train.py --backbone resnet50 --gpu 0 --random-transform pascal datasets/VOC2012 to start training.

Evaluate

  • python3 utils/eval.py to evaluate by specifying model path there.

About

FSAF (Feature Selective Anchor-Free Module for Single-Shot Object Detection) implementation in Keras and Tensorflow

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%