In the beginning I scraped pokemon from trade me then I tried scraping off DeviantArt. Scraping off DeviantArt was a really slow process at first because I think that many of the same images were being downloaded again. To make it faster I looked at the DeviantArt url and saw it goes up in 24 increments so I made a for-loop in the script to do that too.
The pokemon experiments were nice but I thought they were a bit random. I wanted to have some more order in my montage and to flow nicer. So that's when I thought of scraping flowers because it is a more focused set of images. The colourgrid for this was really successful in my opinion with a really nice gradient of colour. The smartgrid was not as good, but still good because it was interesting how it categorised things such as having the sunset photos group, the close up photos group and smaller bunches of flowers group together.
For the next experiment, to create result 2, I scraped food. The reason I did this is because I wanted a data set which is still obvious, but also has a variety of shapes and colours. I think this was really successful in grouping the deserts away from the meals.
For the third result I wanted to experiment what a different layout would look like so I scraped landscape paintings. This was also really successful grouping the brighter, colourful images apart from the darker, duller images.
My final upload is an updated version of the customgrid version of the food scrape. Something to note further about the groupings is that it is quite interesting that the healthy foods group together, but there are also lollies which group near the berries which are not healthy. I think that the food scrape was the most successful because it has more obvious variety with a range of colours and as mentioned earlier it has successful groupings.
Throughout the project the main thing I payed attention to when manually filtering was to get rid of borders of images.