mujiubai / mmdetection-wt

环境安装：

conda create --name mmlab python=3.8 -y
conda activate mmlab
pip install torch==1.10.1+cu111 torchvision==0.11.2+cu111 torchaudio==0.10.1 -f https://download.pytorch.org/whl/cu111/torch_stable.html
pip install tensorboard
pip install -U openmim
mim install mmengine
mim install "mmcv>=2.0.0"
pip install pycocotools
pip install shapely
pip install terminaltables
pip install scipy

分布式训练

CUDA_VISIBLE_DEVICES=0,1 bash tools/dist_train.sh configs/faster_rcnn/faster-rcnn_r50_fpn_1x_coco.py 2

CUDA_VISIBLE_DEVICES=0,1 bash tools/dist_train.sh configs/yolof/yolof_r50-c5_8xb8-1x_coco.py 2

测试

CUDA_VISIBLE_DEVICES=0 python tools/test.py configs/faster_rcnn/faster-rcnn_r50_fpn_1x_coco.py work_dirs/faster-rcnn_r50_fpn_1x_coco/epoch_300.pth --show-dir show_dir

CUDA_VISIBLE_DEVICES=0 python tools/test.py configs/yolof/yolof_r50-c5_8xb8-1x_coco.py work_dirs/yolof_r50-c5_8xb8-1x_coco/epoch_250.pth --show-dir show_dir

问题

list index out of range

# 在配置文件中添加以下代码
class_name=('xyr', 'rmz', 'qf',"zj","hnpt","bt","fzthxm","hhytj")
num_classes=len(class_name)
metainfo=dict(classes=class_name)
# 然后在train_loader设置metainfo，例如：
train_dataloader = dict(
    batch_size=8, num_workers=8, dataset=dict(pipeline=train_pipeline,metainfo=dict(classes=class_name),))

输出特定iou下的recall

#修改安装的pycocotools下cocoeval.py 的_summarizeDets函数

stats[7] = _summarize(0, maxDets=self.params.maxDets[1])

stats[7] = _summarize(0, iouThr=.5, maxDets=self.params.maxDets[1])

更改标注字体大小和框线大小

#修改文件/mnt/data0/home/wangjiangbo/miniconda3/envs/mmlab/lib/python3.8/site-packages/mmdet/visualization/local_visualizer.py
#注意！不是本项目目录下的mmdet，必须是安装的那个，否则无效
def _draw_instances(self, image: np.ndarray, instances: ['InstanceData'],
                        classes: Optional[List[str]],
                        palette: Optional[List[tuple]])
#修改以下函数调用的参数即可
self.draw_bboxes
self.draw_texts                        

绘制图

#faster-rcnn
python tools/analysis_tools/analyze_logs.py plot_curve work_dirs/faster-rcnn_r50_fpn_1x_coco/20240327_215608/vis_data/scalars.json --keys loss loss_rpn_cls loss_rpn_bbox loss_cls loss_bbox --out losses.pdf --legend  loss loss_rpn_cls loss_rpn_bbox loss_cls loss_bbox