PyTorch-SSD has been deprecated. Please see torchcv, which includes an implementation of SSD300/SSD512.

Test

I do not recommend training SSD from scratch. Use pretrained VGG model helps a lot to achieve lower losses.

I use the pretrained pytorch/vision VGG16 model from PyTorch model zoo.

This implementation is initially inspired by:

Name		Name	Last commit message	Last commit date
Latest commit History 38 Commits
image		image
script		script
voc_data		voc_data
LICENSE		LICENSE
README.md		README.md
datagen.py		datagen.py
encoder.py		encoder.py
multibox_layer.py		multibox_layer.py
multibox_loss.py		multibox_loss.py
ssd.py		ssd.py
test.py		test.py
train.py		train.py
utils.py		utils.py