It has been more than a week since I posted any updates on the machine learning workflow. This was because I wanted to get some content to create the next video, and I did not have it readily available.
Here is an example of my training session where I execute training session to identify U.S.A and Texas flags.
With a high +90% confidence that its a flag, one was missing in the picture, but I was pretty happy that it performed this much given the quality of the images.
On this test, nothing was detected. I did have one camera which was stuck in IR mode and I left it on to see if it fixed itself, but it did not. I will be submitting these frames for training too.
Overall, I am pleased so far with the results. I had about 2100 images as part of the dataset for training, and there were no rotations introduce. I still have more data to process, but with this initial data, I can extra more frames from other video and quickly detect the flags.
For more details, please see my upcoming videos on data collection coming up this week.