?????? 市面上暫時還沒有找到可以在消費機(jī)顯卡上實時運行的MaskRCnn，TensorFlow即使是C++版本訓(xùn)練在coco數(shù)據(jù)集上的模型也是慢的要死，最后不堪忍受，只能放棄。

?????? 經(jīng)歷了一些列fuckingDog的復(fù)雜配置之后，終于配置成功了。測試一把

?????? Detectron地址：https://github.com/facebookresearch/Detectron

一、預(yù) 安裝環(huán)境

??????? #查看python ?? ubuntu16.04 默認(rèn)安裝了2.7 和3.5

??????? whereis python

??????? #安裝pip

??????? sudo apt-get install python-pip

??????? #根據(jù)Caffe2的介紹，安裝其他庫

??????? 使用cmake make之時，選擇blas為openblas

???????

??????? sudo apt-get install libopenblas-dev

???????? 省去重新安裝eigen3.3Up

...................................................

?????? #安裝caffe2成功后安裝detection，根據(jù)detection Github上的介紹

...................................................

二、InferImage

把Detectron導(dǎo)入Eclipse ，把infer_sample.py復(fù)制到根目錄;

在debug選項里面添加參數(shù)：

--cfg configs/12_2017_baselines/e2e_mask_rcnn_R-101-FPN_2x.yaml
--output-dir out/? --image-ext? jpg --wts /home/wishchin/build/detectron/model/model_final.pkl demo

結(jié)果輸出到out文件夾

?

對于自己的數(shù)據(jù)，檢測結(jié)果還不錯！

時間消耗在1秒左右

INFO infer_simple.py: 145: Inference time: 0.906s
INFO infer_simple.py: 147:? | im_detect_bbox: 0.765s
INFO infer_simple.py: 147:? | misc_mask: 0.091s
INFO infer_simple.py: 147:? | im_detect_mask: 0.041s
INFO infer_simple.py: 147:? | misc_bbox: 0.008s

NFO infer_simple.py: 111: Processing image/nordic/livingroom (40).jpg -> image/nordicOut/livingroom (40).jpg.pdf
INFO infer_simple.py: 119: Inference time: 0.846s
INFO infer_simple.py: 121:? | im_detect_bbox: 0.783s
INFO infer_simple.py: 121:? | misc_mask: 0.017s
INFO infer_simple.py: 121:? | im_detect_mask: 0.035s
INFO infer_simple.py: 121:? | misc_bbox: 0.010s
INFO infer_simple.py: 124:? \ Note: inference on the first image will be slower than the rest (caches and auto-tuning need to warm up)
INFO infer_simple.py: 111: Processing image/nordic/livingroom (32).jpg -> image/nordicOut/livingroom (32).jpg.pdf
INFO infer_simple.py: 119: Inference time: 0.366s
INFO infer_simple.py: 121:? | im_detect_bbox: 0.320s
INFO infer_simple.py: 121:? | misc_mask: 0.015s
INFO infer_simple.py: 121:? | im_detect_mask: 0.020s
INFO infer_simple.py: 121:? | misc_bbox: 0.011s
INFO infer_simple.py: 111: Processing image/nordic/livingroom (57).jpg -> image/nordicOut/livingroom (57).jpg.pdf
INFO infer_simple.py: 119: Inference time: 0.514s
INFO infer_simple.py: 121:? | im_detect_bbox: 0.445s
INFO infer_simple.py: 121:? | misc_mask: 0.017s
INFO infer_simple.py: 121:? | im_detect_mask: 0.042s
INFO infer_simple.py: 121:? | misc_bbox: 0.011s
INFO infer_simple.py: 111: Processing image/nordic/livingroom (41).jpg -> image/nordicOut/livingroom (41).jpg.pdf
INFO infer_simple.py: 119: Inference time: 0.336s
INFO infer_simple.py: 121:? | im_detect_bbox: 0.297s
INFO infer_simple.py: 121:? | misc_mask: 0.006s
INFO infer_simple.py: 121:? | im_detect_mask: 0.021s
INFO infer_simple.py: 121:? | misc_bbox: 0.012s
INFO infer_simple.py: 111: Processing image/nordic/livingroom (65).jpg -> image/nordicOut/livingroom (65).jpg.pdf
INFO infer_simple.py: 119: Inference time: 0.582s
INFO infer_simple.py: 121:? | im_detect_bbox: 0.483s
INFO infer_simple.py: 121:? | misc_mask: 0.025s
INFO infer_simple.py: 121:? | im_detect_mask: 0.063s
INFO infer_simple.py: 121:? | misc_bbox: 0.011s
INFO infer_simple.py: 111: Processing image/nordic/livingroom (42).jpg -> image/nordicOut/livingroom (42).jpg.pdf
INFO infer_simple.py: 119: Inference time: 0.397s
INFO infer_simple.py: 121:? | im_detect_bbox: 0.348s
INFO infer_simple.py: 121:? | misc_mask: 0.014s
INFO infer_simple.py: 121:? | im_detect_mask: 0.024s
INFO infer_simple.py: 121:? | misc_bbox: 0.011s

總結(jié)

以上是生活随笔為你收集整理的Detectron-MaskRCnn: 用于抠图的FCNN的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。