MMDetection安装

一、MMDetection

下载&＃xff1a;地址

环境&＃xff1a;

• anaconda3
• cuda10.1 (10.0应该也可以)
• python3.6
• torch >&＃61; 1.3
• opencv 4.1

二、MMCV 安装

1&＃xff09;两种版本&＃xff1a;

• mmcv&＃xff1a;精简版&＃xff0c;没有CUDA操作&＃xff0c;但具有所有其他功能&＃xff0c;类似于mmcv <1.0.0。当您不需要那些CUDA操作时&＃xff0c;它很有用。
• mmcv-full&＃xff1a;全面&＃xff0c;具有完整功能和各种CUDA操作。构建需要更长的时间。

2&＃xff09;版本选择&＃xff1a;

3&＃xff09;文档&＃xff1a;地址

4&＃xff09;安装&＃xff08;我是torch 1.6、MMDetection 2.7 &＃xff09;&＃xff1a;

pip install mmcv-full&＃61;&＃61;&＃61;1.1.5 -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.6.0/index.html# pip install mmcv-full&＃61;&＃61;&＃61;{mmcv_version} -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.6.0/index.html# pip install mmcv-full&＃61;&＃61;latest&＃43;torch1.6.0&＃43;cu101 -f https://download.openmmlab.com/mmcv/dist/index.html#其他方式# pip install mmcv
# pip install mmcv-full

PS&＃xff1a;mmcv 与 mmcv-full 不能同时存在

pip uninstall mmcv

5&＃xff09;编译方式安装&＃xff08;若上面方式都不行&＃xff09;&＃xff1a;

git clone https://github.com/open-mmlab/mmcv.git
cd mmcv
MMCV_WITH_OPS&＃61;1 pip install -e . # package mmcv-full will be installed after this step

6&＃xff09;针对 win10 用户&＃xff1a;

•  确保 cl 有效

C:\Users\yuki>cl
用于 x64 的 Microsoft (R) C/C&＃43;&＃43; 优化编译器 19.00.24245 版
版权所有(C) Microsoft Corporation。保留所有权利。用法: cl [ 选项... ] 文件名... [ /link 链接选项... ]

• 若不能调用 环境变量添加&＃xff1a;

添加内容&＃xff08;按实际版本&＃xff09;&＃xff1a;C:\Program Files (x86)\Microsoft Visual Studio\2019\Professional\VC\Tools\MSVC\14.23.28105\bin\Hostx86\x64

三、安装 Cpython

pip install cython # conda install cython

四、安装 MMDetection

git clone https://github.com/open-mmlab/mmdetection.git
cd mmdetectionpip install -r requirements/build.txt
pip install -v -e . # python setup.py develop# pip install -v -e . -i https://mirrors.aliyun.com/pypi/simple

ps:按自己需要选择版本

ps:若win10的话&＃xff0c;管理员 cmd 打开窗口进行操作

五、测试

from argparse import ArgumentParserfrom mmdet.apis import inference_detector, init_detector, show_result_pyplot
import mmcv
import cv2
from scipy import ndimage
import numpy as npdef show_result_ins(img,result,class_names,score_thr&＃61;0.3,sort_by_density&＃61;False,out_file&＃61;None):assert isinstance(class_names, (tuple, list))img &＃61; mmcv.imread(img)img_show &＃61; img.copy()h, w, _ &＃61; img.shapecur_result &＃61; result[0]seg_label &＃61; cur_result[0]seg_label &＃61; seg_label.cpu().numpy().astype(np.uint8)cate_label &＃61; cur_result[1]cate_label &＃61; cate_label.cpu().numpy()score &＃61; cur_result[2].cpu().numpy()vis_inds &＃61; score > score_thrseg_label &＃61; seg_label[vis_inds]num_mask &＃61; seg_label.shape[0]cate_label &＃61; cate_label[vis_inds]cate_score &＃61; score[vis_inds]if sort_by_density:mask_density &＃61; []for idx in range(num_mask):cur_mask &＃61; seg_label[idx, :, :]cur_mask &＃61; mmcv.imresize(cur_mask, (w, h))cur_mask &＃61; (cur_mask > 0.5).astype(np.int32)mask_density.append(cur_mask.sum())orders &＃61; np.argsort(mask_density)seg_label &＃61; seg_label[orders]cate_label &＃61; cate_label[orders]cate_score &＃61; cate_score[orders]np.random.seed(42)color_masks &＃61; [np.random.randint(0, 256, (1, 3), dtype&＃61;np.uint8)for _ in range(num_mask)]for idx in range(num_mask):idx &＃61; -(idx &＃43; 1)cur_mask &＃61; seg_label[idx, :, :]cur_mask &＃61; mmcv.imresize(cur_mask, (w, h))cur_mask &＃61; (cur_mask > 0.5).astype(np.uint8)if cur_mask.sum() &＃61;&＃61; 0:continuecolor_mask &＃61; color_masks[idx]cur_mask_bool &＃61; cur_mask.astype(np.bool)img_show[cur_mask_bool] &＃61; img[cur_mask_bool] * 0.5 &＃43; color_mask * 0.5cur_cate &＃61; cate_label[idx]cur_score &＃61; cate_score[idx]label_text &＃61; class_names[cur_cate]# label_text &＃43;&＃61; &＃39;|{:.02f}&＃39;.format(cur_score)center_y, center_x &＃61; ndimage.measurements.center_of_mass(cur_mask)vis_pos &＃61; (max(int(center_x) - 10, 0), int(center_y))cv2.putText(img_show, label_text, vis_pos,cv2.FONT_HERSHEY_COMPLEX, 0.3, (255, 255, 255)) # greenif out_file is None:return imgelse:mmcv.imwrite(img_show, out_file)def main():parser &＃61; ArgumentParser()parser.add_argument(&＃39;--img&＃39;, help&＃61;&＃39;Image file&＃39;,default&＃61;&＃39;demo.jpg&＃39;)parser.add_argument(&＃39;--config&＃39;, help&＃61;&＃39;Config file&＃39;,default&＃61;&＃39;../configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py&＃39;)parser.add_argument(&＃39;--checkpoint&＃39;, help&＃61;&＃39;Checkpoint file&＃39;,default&＃61;&＃39;../checkpoints/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth&＃39;)parser.add_argument(&＃39;--device&＃39;, default&＃61;&＃39;cuda:0&＃39;, help&＃61;&＃39;Device used for inference&＃39;)parser.add_argument(&＃39;--score-thr&＃39;, type&＃61;float, default&＃61;0.3, help&＃61;&＃39;bbox score threshold&＃39;)args &＃61; parser.parse_args()# build the model from a config file and a checkpoint filemodel &＃61; init_detector(args.config, args.checkpoint, device&＃61;args.device)# test a single imageresult &＃61; inference_detector(model, args.img)# show the resultsshow_result_pyplot(model, args.img, result, score_thr&＃61;args.score_thr)# show_result_ins(args.img, result, model.CLASSES, score_thr&＃61;0.25, out_file&＃61;"demo_out.jpg")if __name__ &＃61;&＃61; &＃39;__main__&＃39;:main()


六、已测试系统

• win10
• ubuntu18.04
• Jetpack4.4