In the paper and in the code, it looks like the PAT tuning is supposed to be for only 1 image x and 1 mask b. Would PATMAT still work for any mask b and any image x of same identity? Table 1 shows that tuning is 65 minutes and testing is 0.04 seconds. But since it only works for 1 image x and 1 mask b, why is the separation between tuning and inference speed even relevant? Thank you for any answers.
In the paper and in the code, it looks like the PAT tuning is supposed to be for only 1 image x and 1 mask b. Would PATMAT still work for any mask b and any image x of same identity? Table 1 shows that tuning is 65 minutes and testing is 0.04 seconds. But since it only works for 1 image x and 1 mask b, why is the separation between tuning and inference speed even relevant? Thank you for any answers.