/pytorch-ssd

SSD: Single Shot multibox Detector - Pytorch

Primary LanguagePython

SSD: Single Shot multibox Detector - Pytorch

Repository: https://github.com/sgrvinod/a-PyTorch-Tutorial-to-Object-Detection
You can run object detection in video using webcam

Paper

SSD: Single Shot MultiBox Detector ECCV 2016
Wei Liu 1, Dragomir Anguelov 2, Dumitru Erhan 3, Christian Szegedy 3, Scott Reed 4,
Cheng-Yang Fu 1, Alexander C. Berg 1 
1 UNC Chapel Hill, 2 Zoox Inc., 3 Google Inc., 4 University of Michigan, Ann-Arbor

Dependencies

  • [Python 3.6]
  • [PyTorch 1.1.0]

Usage

1. Cloning the repository

$ git clone https://github.com/jinyoungHan/pytorch-ssd.git
$ cd pytorch-ssd/

2. Dataset

VOC2012 test
VOC2007 train
VOC2007 test

Extract both the VOC 2007 trainval and 2007 test data to the same location
data/VOC2007/train
data/VOC2007/test

After downloading datasets

$ python data2json.py

3. Training

On a GTX 1060, each epoch of training required about 30 minutes.

$ python train.py

4. Test

i: image , image path or no image path(default path)
w: webcam, no image path

$ python detect.py <i or w> <image path>

Results

1. Epoch 5 - Loss 3.865

2. Epoch 10 - Loss 3.282

3. Epoch 25 - Loss 2.972