mmdetection2.17可视化:pr曲线、FPS Benchmark、anchor、各类loss/mAP、Yolo的anchor

anchor可视化

  • 在目录:mmdet\models\dense_heads\anchor_head就可以可视化anchor,插入位置为: anchor_list, valid_flag_list = self.get_anchors(featmap_sizes, img_metas, device=device)label_channels = self.cls_out_channels if self.use_sigmoid_cls else 1 之间,插入的代码具体如下:
   random_n = 1
   img_shape = img_metas[0]['pad_shape']
   imgh,imgw = img_shape[:

你可能感兴趣的:(视觉检测图像分割干货,mmdetection,可视化,FPS)