I. Description of the txt file in yolo:
Second, yolo runs video and picture file formats:
Third, the reasons for the poor results of yolov5 training:
1. Underfitting:
The phenomenon of poor performance on the training set and poor performance on the test set may be caused by underfitting, due to too much generalization and a high false recognition rate solution:
1) Increase the number of positive samples in the dataset, increase the number of samples of the main features
2) Increase the number of training sessions
3) Reduce the regularization parameter
2. Overfitting.
Performs well on the training set and poorly on the test set (model is too complex) Solution:
1) Increase the number of samples for other features, and retrain the network.
2) Training data is too small a proportion of the total data, increasing the training volume of the data
3. The loss value is no longer small enough to indicate that it has been trained.
IV. yolov5 training results (train file) analysis
1. confusion_matrix.png (confusion matrix)
The confusion matrix summarizes the predictions of the classification problem, showing which parts of the classification model are confused when making predictions.
2. F1_curve：
Relationship between F1-score and confidence level.F1-score (F1-score) is a measure of a classification problem and is the harmonic mean of precision and recall, with a maximum of 1 and a minimum of 0,...1 is the best, 0 is the worst
3.
First graph classes:amount of data per category
Second figure labels: labels
The third figure center xy
The fourth figure labels the length and width of the labels
4. labels_corrrelogram.jpg Not known at this time
5. P_curve.png :
Plot of accuracyprecision and confidenceconfidence
6. PR_curve.png:
The PR curve in theP is for precision.，R stands for recall.In general, recall is set as the horizontal coordinate and precision as the vertical coordinate. the area enclosed under the PR curve is the AP, and the average of the APs of all categories is the Map.
If one of the curves A of the PR graph completely encompasses the curve B of the other learner, it can be asserted that A outperforms BWhen A and B cross, the comparison can be made based on the size of the area under the curve. The general training results are mainly observed in terms of precision and recall fluctuations (if the fluctuations are not very large, then the training is more effective)
- Precision and Recall are often contradictory performance metrics;
- Increase Precision == Increase the threshold for the biclassifier to predict positive examples == Make the positive examples predicted by the biclassifier as true positive examples as possible;
- Increase Recall == Lower the biclassifier's threshold for predicting positive examples == Make the biclassifier pick out as many real positive examples as possible
7. R_curve.png :Relationship between recall and confidence level
8. ：
- Box_loss: YOLO V5 uses GIOU Loss as the loss of the bounding box, Box is presumed to be the mean value of the GIoU loss function, the smaller the box the more accurate;
- Objectness_loss: presumed to be the mean target detection loss, the smaller the target detection the more accurate;
- Classification_loss: presumed to be the mean value of the classification loss, the smaller the classification the more accurate it is;
- Precision: precision (found right positive class/all found positive classes);
- Recall: true for POSITIVE accuracy, i.e., how many positive samples were found (how many were recalled).Recall describes how many true positive examples in the test set were picked by the binary classifier from the perspective of true results, i.e., how many true positive examples were recalled by that binary classifier.
- val Box_loss: validation set bounding box loss.
- val Objectness_loss: validation set target detection loss mean value.
- val classification_loss: validation set classification loss mean.
- [email protected]:.95 (mAP@[.5:.95]): denotes the average mAP at different IoU thresholds (from 0.5 to 0.95 in steps of 0.05) (0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.8, 0.85, 0.9, 0.95). [email protected]: denotes the average mAP for thresholds greater than 0.5. Then [email protected] & [email protected]:0.95 Evaluation of training results. mAP is the area enclosed after plotting with Precision and Recall as the two axes. m denotes the average, the number after @ denotes the threshold for determining iou as a positive or negative sample, @0.5:0.95 denotes that the threshold is taken to be the mean value after taking 0.5:0.05:0.95
Note: The above information and images are from the official website of YOLOV5, excellent authors of CSDN and their own trained datasets, infringement deleted.
I am learning about event camera detection and other things (white guy), and would like to learn and share with the many great people who are learning about event cameras!
summarize
To this article on yolov5 some simple description of the article is introduced to this, more related yolov5 txt file, training results analysis content please search for my previous articles or continue to browse the following related articles I hope you will support me in the future!