functions: pva-pytorch using pytorch-1.0 add multi-scales image train

pytorch1.0,opencv-python,skimage

step1: install git clone https://github.com/syshensyshen/pva-pytorch.git

'cd pva-pytorch/lib' python setup.py build develop

step2: train CUDA_VISIBLE_DEVICES=1 python trainval_net.py --batch_size 8
--save_dir ./save_models
--xml_path /data/datasets/defect_detection/voc/VOC2007/Annotations/
--img_path /data/datasets/defect_detection/voc/VOC2007/JPEGImages/
--network pva --classes 6

pretrainedmode：链接：https://pan.baidu.com/s/1eLdUheV3FHmsyYs-MQLKBQ 提取码：psnw

reference code: jwyang/faster-rcnn.pytorch sanghoon/pytorch_imagenet