New-SSD
High precision and fast speed for SSD
| Model | VOC2007 test | VOC2012 test | FPS (TitanX) | #Boxes | Input resolution |
|---|---|---|---|---|---|
| SSD300 (VGG16) | 77.2 | 75.8 | 46 | 8732 | 300 x 300 |
| SSD512 (VGG16) | 79.8 | 78.5 | 19 | 24564 | 512 x 512 |
| SSD300 (VGG16)(this repo) | 80.0 | 77.5 | - | 8732 | 300 x 300 |
| SSD321 (ResNet101)(this repo) | 79.5 | 76.4 | 27.5 | 10325 | 321 x 321 |
| SSD512 (VGG16)(this repo) | 81.6 | 80.0 | - | 24564 | 512 x 512 |
| SSD513 (ResNet101)(this repo) | 83.0 | 81.3 | 16 | 25844 | 513 x 513 |
Dependencies
Library: OpenCV-Python, PyTorch>0.4.0, Ubuntu 14.04
Dataset
PascalVOC
# Download the data.
cd $HOME/data
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtrainval_06-Nov-2007.tar
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar
# Extract the data.
tar -xvf VOCtrainval_11-May-2012.tar
tar -xvf VOCtrainval_06-Nov-2007.tar
tar -xvf VOCtest_06-Nov-2007.tarMSCOCO 2017 (download link)
#step1: download the following data and annotation
2017 Train images [118K/18GB]
2017 Val images [5K/1GB]
2017 Test images [41K/6GB]
2017 Train/Val annotations [241MB]
#step2: arrange the data to the following structure
COCO
---train
---test
---val
---annotationsTrain/Test/Evaluation
1. Change the mode in main.py
2. Change parameters such as root (data directory) in config/ssd_config.py
3. python main.py