projectayre / ayre

Visual Question Answering with added novel Semantic Analysis approach. Inspired in part by Armored Core 6: Fires of Rubicon.

3 stars 2 forks source link

Sentiment analysis #4

Closed hauntedcupoftea closed 1 year ago

hauntedcupoftea commented 1 year ago

Selected this dataset: https://datasets.simula.no/image-sentiment/ for now. It has the most detailed emotional information we have seen so far. However, this dataset can prove biased because all images are related to disasters.

7Zenox commented 1 year ago

These are worth exploring:

hauntedcupoftea commented 1 year ago

8 has a valid solution, so this isn't really needed. We need to get to actually making the project, and I doubt we'll find a dataset that covers the range of emotions as well as this. Reopn if you can prove me wrong (please do, anything better is always nicer)

hauntedcupoftea commented 1 year ago

Reopening this because @7Zenox and @Dekode1859 want to potentially replace the dataset with something else. What we have discussed so far:

Current dataset is small and niche, and has a low of bias against "general" and "social media" data.
However, from a totally research based stand, pre-processing the 26 output labels results in a more novel approach that we can add to academic research
Someone was talking about alt datasets and their importance, as well as potential for better output. Can't remember who, but please just talk this out and find a fix because I'm slowly losing patience with all my work.

hauntedcupoftea commented 1 year ago

This task will now be done via a modification of the image semantic segmentation. Had a small eureka moment.