Originally published at: Deep Learning for Classifying Hotel Aesthetics Photos | NVIDIA Technical Blog
Most people love to travel. The desire to experience something unfamiliar, such as learning about a new culture or meeting different people, is something that probably all of us have felt in our lives. However, travel means making travel arrangements, which usually includes booking a hotel room. Finding a hotel room is however not always…
Very cool project. Does EML require many votes per image to be accurate? I imagine one would need more votes per image for using EML compared to regressing on the means. Eg, how many votes did you collect for your dataset composed of 1k images?
It would be interesting to commission a crowd sourced labeling task where you give the annotators pairs of images A & B and they are to make a relative judgement on which image is better for quality and aesthetics. This might have an advantage over your point-wise likert scale style approach as you will probably have less disagreement between annotators: cleaner data.