PASCAL VOC2007 Test
Scores are mean Average Precision (mAP) with PASCAL VOC2007 metric.
COCO Test
Detect objects in an given image. (Please download pre-trained model to ~/.torch/models
first. --- If you put pre-trained model to other folder, please change the --root
)
$ python demo_ssd_cv.py [--network ssd_300_vgg16_astrous_voc] [--images <image>.jpg] [--cuda]
Note:please choose one of the model listed in performance as network. There are several images in
../png
, you can choose one as demo
The default data root is ~/.torch/datasets
(You can build a soft-link to it)
$ python [--network ssd_300_vgg16_astrous_voc] [--data-shape 300|512] [--batch-size 8] [--dataset voc|coco] [--cuda] [--root pretrained-model folder]
Note:
- please make sure the network and data-shape is consistent.
- the default root is
~/.torch/models
(And make sure the pre-trained model is named as<--network>.pth
)
Download pre-trained backbone and put it on ~/.torch/models
Recommend to using distributed training.
$ export NGPUS=4
$ python -m torch.distributed.launch --nproc_per_node=$NGPUS train_ssd_cv.py [--network vgg16_atrous] [--data-shape 300] [--dataset voc|coco] [--batch-size 32] [--test-batch-size 16] [--lr 1e-3] [--lr-decay-epoch 160,200] [--lr-decay 0.1] [--epochs 240] [-j 16] [--lr-mode step|cos] [--warmup-factor 0.01] [--log-step 10]
Note:
- the batch-size is per-batch-size-in-one-gpu
Pre-trained Backbone
darknet53 | mobilenet1.0 |
---|---|
BaiduYun/Google Drive | BaiduYun/Google Drive |