IMG_8720 scale(0.50, 0.60, +0.05) FP(0) FN(0)

Group 1 scale(0.50, 0.60, +0.05) FP(1) FN(0)

The false positive has a much higher MSE than the others but the deviation from facial mean and general colors are close enough to scrape by the thresholds.

Group 2 scale(0.50, 0.60, +0.05) FP(0) FN(0)

Group 3 scale(0.50, 0.60, +0.05) FP(0) FN(0)

Group 5 scale(0.50, 0.60, +0.05) FP(0) FN(0)

Previous Group 1 scale(0.40, 0.80, +0.02) FP(0) FN(0)

Previous Group 2 scale(0.40, 0.80, +0.02) FP(1) FN(1)

The false positive is still skin so its close. The false negative probably has something to do with the fact that his head is tilted away from the camera.

Friends in Vegas scale(0.40, 0.80, +0.02) FP(1) FN(3)

The neck apparently had a better overall error and it overlapped with the face which resulted in one false positive and one false negative. The other two false negatives are harder to reason about, possibly head tilt issues but the far right face is found. I was able to find the large face in a previous iteration but upon tightening some of the thresholds, I lost it.

More Friends in Vegas scale(0.40, 0.80, +0.02) FP(0) FN(2)

This is a special case. Not quite human face. :)

Picture from UW CSE Site scale(0.40, 0.80, +0.02) FP(2) FN(1)

The false negative is hard to understand. The shirt false negative has some detail and falls below the color threshold. As for the brick, it has a decent MSE and is relatively close to the facial mean... unfortunatly.