Open nguyenpbui opened 3 months ago
Hello authors,
Thank you for the awesome work.
In your implementation, you set the upper threshold to 0.98 and the lower one to 0.97.
Can you explain your choice or any advice on the best thresholds?
Thank you very much.
If we set these thresholds too high (0.98 and 0.97), the proposed SAT idea seems similar to conventional contrastive learning since all other images are considered negative samples.
Hello authors,
Thank you for the awesome work.
In your implementation, you set the upper threshold to 0.98 and the lower one to 0.97.
Can you explain your choice or any advice on the best thresholds?
Thank you very much.