Faster RCNN实现（2）

环境 ubantu16.04+cudnn7.0+cuda_9.0.176
Pytorch1.0+Tensorflow-gpu 1.9.0+python3.6.5+ anaconda3
先下载https://github.com/jwyang/faster-
rcnn.pytorch/branches

一、数据准备和预训练权重下载

参考（1）https://blog.csdn.net/yx868yx/article/details/105550545

apple和peach数据集、vgg16_caffe.pth权重

二、训练

1、先编译：

$ cd lib
$ python setup.py build develop

2、训练

$ CUDA_VISIBLE_DEVICES=0 python trainval_net.py --dataset pascal_voc --net vgg16 --bs 16  --cuda

报错:ModuleNotFoundError

Traceback (most recent call last):
File “trainval_net.py”, line 27, in
from roi_data_layer.roidb import combined_roidb
File “/home/yuxin/faster-rcnn.pytorch-pytorch-1.0/lib/roi_data_layer/roidb.py”, line 9, in
from datasets.factory import get_imdb
ModuleNotFoundError: No module named ‘datasets.factory’

solution:

$ export PYTHONPATH=/home/yuxin/faster-rcnn.pytorch-pytorch-1.0/lib

报错：ImportError
用到COCO数据集不安装COCOAPI时会报错

Traceback (most recent call last):
File “trainval_net.py”, line 27, in
from roi_data_layer.roidb import combined_roidb
File “/home/yuxin/faster-rcnn.pytorch-pytorch-1.0/lib/roi_data_layer/roidb.py”, line 9, in
from datasets.factory import get_imdb
File “/home/yuxin/faster-rcnn.pytorch-pytorch-1.0/lib/datasets/factory.py”, line 15, in
from datasets.coco import coco
File “/home/yuxin/faster-rcnn.pytorch-pytorch-1.0/lib/datasets/coco.py”, line 23, in
from pycocotools.coco import COCO
File “/home/yuxin/faster-rcnn.pytorch-pytorch-1.0/lib/pycocotools/coco.py”, line 60, in
from . import mask
File “/home/yuxin/faster-rcnn.pytorch-pytorch-1.0/lib/pycocotools/mask.py”, line 3, in
from . import _mask
ImportError: cannot import name ‘_mask’

solution:

$ cd data
$ git clone https://github.com/pdollar/coco.git 
$ cd coco/PythonAPI
$ make
$ cd ../../..

三、测试

命令:

$ python test_net.py --dataset pascal_voc --net vgg16 --checksession 1 --checkepoch 20 --checkpoint 286  --cuda

四、测试demo结果

$ python demo.py --net vgg16  --checksession 1  --checkepoch 20 --checkpoint 286 --cuda --load_dir models

五、视频测试demo

命令：

$ python demo.py --net vgg16  --checksession 1  --checkepoch 20 --checkpoint 286 --cuda --load_dir models --webcam 0

检测结果过拟合了！