An interactive visualisation of semantically similar movie trailers.
Visit the page here
Visualizing.Image.Fields.-.Google.Chrome.2021-10-03.16-05-45.mp4
A visualisation of the MMX-20 Dataset and results from the paper
E. Fish, J. Weinbren and A. Gilbert, "Rethinking Genre Classification With Fine Grained Semantic Clustering," 2021 IEEE International Conference on Image Processing (ICIP), 2021, pp. 1274-1278, doi: 10.1109/ICIP42928.2021.9506751.
The page features ~8000 movie trailer embeddings which have been extracted using the model architecture described in the paper above. All embeddings are clustered using only audio-visual information without metadata, although you can select metadata from the drop down menus.
Both T-SNE and UMAP projections are available.
The embedding visualisation is a modified version of the output from PixPlt (Yale and Google Arts) project. Additional features:
- Mouseover images for details
- Clicking on images searches and retrieves video via YouTube API
- Meta Data views and drop downs
- Improved highlighting of selected values
- Node.js added for easy deployment
Note: Due to YoutubeAPI limits, when you select an image the video may not play correctly if the API limit is exceeded.