why the feature extraction is so inaccurate?

[Juruige]

Hello, I tried to run my own data with your program. I used the distance matrix to calculate the distance between each category and the distance between each class sample, but found that the feature extraction is very inaccurate, the extracted vector is irregular.I want to ask you where is the mistake?

[r3a1d3a1]

It downsamples images to 224x224 pixels which is only good enough to recognize one big subject like horse, human, car,... not any sort of details. Methods mentioned in here may fare better, as they usually don't downsample past 500x500 which is still small, but at least 4 times bigger than 224x224.