WSDDN PyTorch

This repository contains the PyTorch implementation of paper Weakly Supervised Deep Detection Networks (CVPR 2016).

• Implementation differences: Spatial regulariser isn't added
• Related code: OICR-PyTorch

Architecture

Results

VOC2007 test

	aero	bike	bird	boat	bottle	bus	car	cat	chair	cow	table	dog	horse	mbike	person	plant	sheep	sofa	train	tv	mAP
Ap	41.4	44.2	25.8	12.2	10.9	56.3	39.0	27.8	7.0	25.5	16.6	24.4	18.0	51.5	1.1	11.2	17.9	10.6	48.0	42.2	26.6

VOC2007 trainval

	aero	bike	bird	boat	bottle	bus	car	cat	chair	cow	table	dog	horse	mbike	person	plant	sheep	sofa	train	tv	mean
CorLoc	68.8	62.4	54.7	33.5	39.7	68.0	67.7	49.1	35.3	54.8	33.8	38.8	37.4	77.1	7.54	43.2	60.8	24.5	68.8	63.1	49.5

Contents

1. Requirements: software
2. Requirements: hardware
3. Basic installation
4. Installation for training and testing
5. Extra Downloads (selective search)
6. Extra Downloads (ImageNet models)
7. Usage

Requirements: software

Python3 packages and versions used (listed using freeze frin pip) are in requirements.txt.

You can create a new virtual environment and then install thses packages from requirements.txt.

conda create -n env_name python=3.6
pip install -r $WSDDN_ROOT/requirements.txt

You can also install these packages by yourself.

Besides, you should install Octave, which is mostly compatible with MATLAB.

sudo apt-get install octave

Requirements: hardware

• I used cuda 9.0 and cudnn 7.0 on Ubuntu 16.04
  • We used an Nvidia GeForce GTX with 10.9G of memory. But it shold be ok to train if you have a GPU with at least 8Gb.
  • NOTICE: different versions of Pytorch have different memory usages.

Basic installation

Clone this repository

git clone https://github.com/CatOneTwo/WSDDN-PyTorch

Installation for training and testing

1. Create a "data" folder in $WSDDN_ROOT and enter in this folder

   cd $WSDDN_ROOT
   mkdir data
   cd data

2. Download the training, validation, test data, and VOCdevkit

   wget https://pjreddie.com/media/files/VOCtrainval_06-Nov-2007.tar
   wget https://pjreddie.com/media/files/VOCtest_06-Nov-2007.tar

3. Extract all of these tars into one directory named VOCdevkit

   tar xvf VOCtrainval_06-Nov-2007.tar
   tar xvf VOCtest_06-Nov-2007.tar

4. Download the VOCdevkit evaluation code adapted to octave

   wget http://inf.ufrgs.br/~lfazeni/CVPR_deepvision2020/VOCeval_octave.tar

5. Extract VOCeval_octave

   tar xvf VOCeval_octave.tar

6. Download pascal annotations in the COCO format

   wget http://inf.ufrgs.br/~lfazeni/CVPR_deepvision2020/coco_annotations_VOC.tar

7. Extract the annotations

   tar xvf coco_annotations_VOC.tar

8. It should have this basic structure

   $VOC2007/                           
   $VOC2007/annotations
   $VOC2007/JPEGImages
   $VOC2007/VOCdevkit        
   # ... and several other directories ...

9. [Optional] download and extract PASCAL VOC 2012.

   wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar
   tar xvf VOCtrainval_11-May-2012.tar

   Observation: The 2012 test set is only available in the PASCAL VOC Evaluation Server to download. You must create a user and download it by yourself. After downloading, you can extract it in the data folder.

Extra Downloads (selective search)

1. Download the proposals data generated by selective search to data folder

   wget http://inf.ufrgs.br/~lfazeni/CVPR_deepvision2020/selective_search_data.tar

2. Extract the proposals

   tar xvf selective_search_data.tar

Extra Downloads (ImageNet models)

1. Download the pre-trained VGG16 model to data folder

   wget http://inf.ufrgs.br/~lfazeni/CVPR_deepvision2020/pretrained_model.tar

2. Extract the pre-trained VGG16 model

   tar xvf pretrained_model.tar

Usage

To Train the WSDDN network on VOC 2007 trainval set:

python3 code/tasks/train.py --cfg configs/baselines/vgg16_voc2007.yaml --model midn

To Evaluate the WSDDN network on VOC 2007:

On trainval (corloc)

 python3 code/tasks/test.py --cfg configs/baselines/vgg16_voc2007.yaml --dataset voc2007trainval --model midn --load_ckpt snapshots/midn/<some-running-date-time>/ckpt/model_step24999.pth

On test (detection mAP)

python3 code/tasks/test.py --cfg configs/baselines/vgg16_voc2007.yaml  --dataset voc2007test --model midn --load_ckpt snapshots/midn/<some-running-date-time>/ckpt/model_step24999.pth

To Visualize the detection results:

After evaluating WSDDN on test dataset, you will get detections.pkl. Then you can run the visualization script to show the results in a openCV window.

python3 code/tasks/visualize.py --cfg configs/baselines/vgg16_voc2007.yaml --dataset voc2007test --detections snapshots/midn/<some-running-date-time>/test/model_step24999/detections.pkl 

You can also save the visualizations as images. First create a folder to save the outputs and pass it with the --output argument

mkdir output    
python3 code/tasks/visualize.py --cfg configs/baselines/vgg16_voc2007.yaml --dataset voc2007test --detections snapshots/midn/<some-running-date-time>/test/model_step24999/detections.pkl --output output 

My model

You can download my model by baidu netdisk:

Link: https://pan.baidu.com/s/1pEpmh4AYgPxW5hVBkJ7Kfw Code: zcn5

Put this folder in your $WSDDN_ROOT, then test and visualize model.

Note <some-running-date-time> is replaced by final in this folder, so you should also make changes when you run the code.

---

Notes

Below is the code structure

• code
  • datasets: VOC dataset file
  • layers: layer and loss files
  • models: WSDDN model based on layers
  • tasks: train, test and visualize files
  • utils: files used for other directories
• configs
  • baselines: config files for model and dataset
• data
  • pretrained_model: VGG16 model weights
  • selective_search_data: selective search for VOC data
  • VOCdevikit: VOC dataset

All code files are in code directory. If you want to design a model based on WSDDN, you can modify layers and model.

References

• Boosted-OICR
• OICR