1 min readMay 28, 2019
- “ the conditions in shoe dataset cannot represent the whole of the visual features”, Yes the variation in the shoe dataset, e.g gender, doesn’t cover all possible visual variation on all datasets
- “half of the dimensions of the features in CNN are not important to distinguish images”, Yes there is a chance some features are redundant or useless.
- I don’t understand what do you mean by “adding loss (CNN) == (max of the whole conditional features) could be possible”
