Skip to content

leowangzi/LightHeadRCNN

Repository files navigation

FasterRCNN

Renferences

Prerequisites

  • Python 2.7 or 3.6
  • Pytorch 1.1 or higher
  • CUDA 9.0 or higher
  • tensorboardX

First of all, clone the code

git clone https://github.com/leowangzi/LightHeadRCNN.git
cd LightHeadRCNN

Then, create a folder:

mkdir data

or

ln -s [source_data] data

What we are doing and going to do

  • Support pytorch-1.1 (master).
  • Support torchvision-0.3 (master).

Benchmarking

We benchmark our code thoroughly on pascal voc datasets, using resnet101 network architecture. Below are the results:

1). PASCAL VOC 2007 (Train/Test: 07trainval/07test, scale=600, ROI Align)

model #GPUs batch size lr lr_decay max_epoch time/epoch mem/GPU mAP
[Res-101] 1 1 1e-3 5 10 0.88 hr 5965 MB 75.06

2). PASCAL VOC 2007&2012 (Train/Test: 07+12trainval/07test, scale=600, ROI Align)

model #GPUs batch size lr lr_decay max_epoch time/epoch mem/GPU mAP
[Res-101] 1 1 1e-3 5 10 0.88 hr 3200 MB 79.80

Releases

No releases published

Packages

No packages published