Skip to content

qqadssp/DSOD-Pytorch

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

13 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DSOD-Pytorch

This is an implementation of DSOD in Pytorch. It is based on the code dsod.pytorch and torchcv

Origin implementation is here and here is the paper

I can train it on PASCAL VOC dataset and the loss also converges, but I am not sure it can achive the same scores as in the paper. Some more modifications need to be done.

Requirment

python 2.7
pytorch 0.4
visdom

Train on VOC

  1. Download this repo
git clone git@github.com:qqadssp/DSOD-Pytorch  
cd DSOD-Pytorch  
  1. Download Pascal VOC dataset and unzip it, its path should be {root_dir}/VOCdevkit

  2. Modify opt.train_img_root in torchcv/utils/config.py with proper img_path

  3. Start visdom server and begin to train

python -m visdom.server  
python train.py main  

Eval

  1. After training some epochs, checkpoint will be saved with name 'dsod.pth' or '##.pth' like '39.pth'. Modify opt.load_path in config.py with 'checkpoint/dsod.pth' or 'checkpoint/39.pth'

  2. Download Pascal VOC testset, modify opt.eval_img_root in config.py with proper path

  3. Evaluate the model

python eval.py

Demo

Training dataset is used here. If you have trained some epochs and get checkpoint, modify opt.load_path and run

python demo.py

Releases

No releases published

Packages

No packages published

Languages