关于目标检测的变种,
1,旋转角度
Train with rotated bounding boxes - will require some theta value in the Label file I guess Detect the rotated boxes. Basically as discussed here for YV3 https://github.com/ultralytics/yolov3/issues/345
There is this: https://github.com/ming71/rotate-yolov3 This is next level: https://github.com/maudzung/Complex-YOLOv4-Pytorch Then: https://github.com/mrlooi/rotated_maskrcnn
2,多边型,并且衍生为分割模型。
poly yolo
3,优化点,模型内进行resize。