![]() Words such as “thought,” “dead,” and “struggle” definitely go more with the themes of indie songs. Indie songs tend to be sadder and introspective, so it follows that words like “yeah,” “oh,” and “hey” don’t appear often, or at all, in the lyrics. This insight is, by far, my favorite, especially in the context of lyrics. The default display shows the full sample, shown below. Hovering over an image lets you view its associated class, so you can use that to see whether the images are grouped in a way you would expect. Filters allow you to narrow the display by class, and zoom controls allow you to zoom in (and out) on the display. With Image Embeddings, you can view projections of images in two dimensions to see the visual similarity between a subset of images and help identify outliers. The bottom blue is very similar to the indie cover in the top left corner, it doesn’t have as vibrant colors, but the faces and figures are much more distinct and clear. The top blue is an example of a rock album with concept art, so it might seem to fit indie more, but the bright blue, purple, and orange coloring doesn’t fit. The pop albums, highlighted in red, might seem to fit indie more because the figure isn’t in a portrait style, but the bright pink and green colors don’t follow the indie palette. ![]() The rock and pop covers have more saturated or deep colors. The colors in these are very muted, and even though two have figures, they are faded and hard to make out. Again, I highlighted in orange some of the indie cover art. ![]() Black and White Activation Maps from Full Sampleįor some cover art, the activation maps show us how the model focuses on color and clarity. I also copied the track’s lyrics from the web.įigure 8. If you are interested in the specifics of how I used the API, you can check that out here. This data included the title of the track and the title and the cover art of the track’s album. I leveraged the Spotify API to collect metadata on 277 songs that fell into the three genres, indie, rock, and pop. In this post, I will show you how I built this model and what it teaches us about the role a record’s cover plays in categorizing and placing an artist’s work into a musical context. As a starting place, I was curious if machine learning could accurately predict an album’s genre from the cover art. Browsing through a collection of images takes a lot less time than listening to clips of songs. This post is meant to be an enjoyable and unique way to explore Visual AI.Īs someone who spends hours searching for new music, getting lost in rabbit holes of ‘related artists’ or ‘you may also like’ tabs, I wanted to see if cover art improves the efficiency of the search process. DataRobot Success Stories See how organizations like yours have realized more value from their AI initiatives. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |