Since the data is enhanced during training, does it make sense to augment it in order to increase the amount of data?
I would like to ask. According to the introduction of yolov5 data set, it is suggested not to enhance the picture, because it will be carried out in the training process. However, when I saw some papers, due to insufficient data, Do some enhancements increase the number of datasets, does the increase in volume affect the quality? The question I struggle with is if you augment the data during training, does it make sense to augment it to increase the amount of data?
0 Answer
Since the data is enhanced during training, does it make sense to augment it in order to increase the amount of data?
I struggle with the point that since the data is enhanced during training, does it make sense to augment it to increase the amount of data?
the reference answer GPT ᴼ ᴾ ᴱ ᴺ ᴬ ᴵ < br / > to the question of the data set to enhance this is a need to according to the specific circumstances of problem. When performing data enhancement, you need to weigh the advantages and disadvantages of the enhancement.
-
On the one hand, data enhancement can increase the diversity of the data set, reduce the risk of overfitting, and improve the generalization ability of the model. On the other hand, enhancement may affect the quality of data set, for example, excessive enhancement may lead to image distortion, information loss and other problems.
-
For YOLOv5 and other target detection models, some enhancement operations(such as random clipping, rotation, scaling, etc.) may cause certain changes in the position and size of target objects, so that the model can better adapt to a variety of different scenes and target objects.
-
Therefore, whether or not to do data enhancement and how to do it depends on the situation. If the data set is already sufficiently diverse, the enhancement operation may have a negative impact on the model training effectiveness. However, if the data set is relatively simple, the enhancement operation can effectively increase the diversity of data and improve the generalization ability of the model.
这家伙很懒，什么都没留下...
