Object Detection Using YOLO v3 Deep Learning, Evaluate Model

In the example of Object Detection Using YOLO v3 Deep Learning, only the code in the picture above is replaced, the rest of the code is the same.

This is the trained Detector.

In the Evaluate Model part, the value of results does not come out.

How can I solve this?

Answer 1

[Vivek Akkala]

It is possible that the network may not have been trained entirely for the provided dataset. Please ensure that the loss function converges to the lowest possible level. To better comprehend loss convergence, you can test the example without modifying the data. Additionally, consider experimenting with different options such as tuning hyper-parameters, adjusting the network architecture, augmenting your data, or adding more training data to improve model performance.

[준석 김]

When I looked at lossinfo, only objLoss was high at 9.9643, and the rest of the loss did not exceed 1.

Will reducing objLoss solve it?

And what is the solution to reduce objLoss?

[Vivek Akkala]

ObjLoss of 9.9 is pretty high. There are high chances that reducing objLoss would solve the problem. Try reducing the objLoss to less than 0.5.

It's difficult to suggest the a solution for reducing the objLoss as it highly depends on data and network architecture. I would start with the options mentioned in Deep Learning Tips and Tricks.