1. Explore ImageNet. ImageNet sample images, Kaggle ImageNet Mini 1000, What surprises you about this data set? What questions do you have? Thinking back to last week’s assignment, can you think of any ethical considerations around how this data was collected Are there privacy considerations with the data?

I was very surprised when I saw these data sets, and I wondered how they could collect so much data. Taking a database of various fish as an example, I wanted to know where the data came from, whether it was uploaded online or searched, and how so much picture data was classified and how these models were trained.

I opened the link and began to look at the images one by one. I found that although the subject content of the images was the same, the environment was very different. Some pictures had clean backgrounds, and some pictures were very messy. I don't think the data in its current form is the best, and the quality of the data needs to be improved, because some of the photos are even dominated by humans, and it's hard for me to see what the fish look like, which sounds ridiculous. The point of collecting data is lost when people can't identify the real subject. I think we should collect more data with clean background and clear picture.

Of course, there are some problems with the data. Some photos show human hands, such as humans catching fish with their hands, and some photos even show human faces (there are many photos), I don't know if this is allowed, I think it involves personal privacy issues. From these pictures, I also found that the database may involve important issues, that is, privacy, the faces involved in these databases, whether they know the content, whether they have permission, and whether users have the right to know who these people are. I think it is necessary to protect people's privacy in the process of collecting data.

1. Using the ml5.js examples above, try running image classification on a variety of images. Pick at least 10 objects in your room. How many of these does it recognize? What other aspects of the image affect the classification, including but not limited to position, scale, lighting, etc.

I don't know if it is because I have been holding the items that need to be sorted by hand, so it is difficult to identify the items, and only 3 items have been identified. A necklace, a pen and a cup. Of the identified items, I found that the closer I held the items to the computer screen, the higher my confidence. Similarly, when I changed the background to a white wall, my confidence increased.

But among the items that were not identified, I found that those items were misidentified because they had some characteristics. For example, my key, because of its length and because I hold it in my hand, has always recognized that I am holding a pistol. What's even more interesting is that I took a dollar bill with a picture of a person on it, so it has always been recognized as a comic book. I think we can't just classify things according to some characteristics, we need to make them more accurate.