Hi,
I use my own dataset.
A total of eight categories,
832 training datasets (104 per category) and 360 validation datasets (45 per category).
The result of the pretext:
The top result of the kNN evaluation is 67.5
Accuracy of top-10 nearest neighbors on train set is 67.9
Accuracy of top-5 nearest neighbors on val set is 76.83
The result of SCAN ACC=0.631, and The result of Self-label ACC=0.3
I used to change the Parameter of the Pretext:
Augmentations, batch size, features_dim, temperaturee
The results of the Pretext are about the same.
And change the Parameter of the SCAN:
batch size, entropy_weight, num_neighbors
I haven't seen a significant improvement.
What should I do to make the result better?
Hi, I use my own dataset. A total of eight categories, 832 training datasets (104 per category) and 360 validation datasets (45 per category).
The result of the pretext:
The result of SCAN ACC=0.631, and The result of Self-label ACC=0.3
I used to change the Parameter of the Pretext: Augmentations, batch size, features_dim, temperaturee The results of the Pretext are about the same.
And change the Parameter of the SCAN: batch size, entropy_weight, num_neighbors I haven't seen a significant improvement. What should I do to make the result better?
Thanks Kelly