HUAWEI CLOUD ModelArts First Experience - Aircraft Identification

I. Introduction

I recently learned about HUAWEI CLOUD's ModelArts, which is a one-stop AI development platform for developers. It provides massive data preprocessing and semi-automatic labeling, large-scale distributed training, automated model generation, and end-to-end for machine learning and deep learning. On-demand deployment of edge-cloud models helps users quickly create and deploy models and manage full-cycle AI workflows.


Follow the guidance of the official documentation: ModelArts usage documentation Successfully created the Yunbao prediction model in the case and performed the prediction operation, and the accuracy rate is quite high.
The recognition effect is also good in the case of multiple images:
insert image description here


Passing pictures similar to Yunbao (Misue Bingcheng-Snow King) will still have a high degree of similarity:


insert image description here


Such a situation may be the reason why the accuracy of the previous data annotation is not high:
insert image description here


So I found some pictures again, relabeled, trained and tried again.


2. Actual combat

1. Model training

Here I found 26 pictures of airliners, and the training pictures have been uploaded to the AI ​​Gallery community: Picture address :
insert image description here


Then label them one by one, similar to the multi-object situation in the following figure, which is also multi-labeled:
insert image description here


The training is completed in about 12 minutes, and the accuracy rate given is 91%:
insert image description here


2. Predict the effect

1) Airplane pictures

Normal airplane pictures:

Predicted similarity: 93%


insert image description here


Predicted similarity: 98%


insert image description here


Darker pictures of planes:

Predicted similarity: 79%


insert image description here


Helicopter pictures:

Predicted similarity: 93%


insert image description here


Fighter aircraft pictures:

Prediction similarity: 98% (the uploading of fighter pictures is always illegal, I don’t understand it; here is the prediction result)


insert image description here


2) Non-aircraft pictures

mascot:

Predicted similarity: 36%


insert image description here


car:

Predicted similarity: 33%


insert image description here


pet:

The prediction result is empty


insert image description here


3. Summary

I am still satisfied with the result when there are only 26 samples. I am using the free version, and I have not set some parameters specially. It may be better to set these results.


In addition, HUAWEI CLOUD's AI Gallery is a HUAWEI CLOUD AI knowledge & training community, which can help you become an AI developer from 0 to 1!

There are many courses and practices for your learning and reference:
insert image description here


Guess you like

Origin blog.csdn.net/momoda118/article/details/122575345